TRANSITIVITY:

TOWARDS A COMPREHENSIVE TYPOLOGY

Seppo Kittilä
PREFACE

Transitivity as a linguistic phenomenon has been a topic of numerous studies. The

present study adds another (and certainly not the last) entry to this list. My first
encounter with the notion of transitivity was some 14 years ago, when I began to
learn German in high school. One of the rules of German grammar stated something
like ‘intransitive verbs expressing motion or change-of-state take sein as their
auxiliary verb in compound past tenses’. The use of such complicated formulations
seemed frustrating to a 16-year-old. I would have certainly laughed, if someone had
told me that in some 10 years I would be working on a PhD thesis on the very notion
of linguistic transitivity. This is, however, what happened. I was seduced by this
fascinating topic, when I had to find a theme for a paper to be presented in the
seminar of general linguistics in the spring of 1997. For some reason I opted for
discussing passivization and typology of passives. To others it seemed bizarre that
someone can be interested in something that insignificant as the typology of
transitivity. As noted above, I would have been one of the people wondering about
this some 10 years back, but obviously this is not the case anymore. The paper
presented at the seminar was later extended, since it served as the basis for my MA
thesis in linguistics. During the writing process of my MA theses I (accidentally at
first) made myself familiar with different aspects of transitivity which finally
resulted in the present study. Four years is a long period of time and only some of
my original ideas are presented on the next 400 pages. Spending four years studying
transitivity from different perspectives has revealed numerous aspects that sorely
need to be studied in detail and I hope that I will able to focus on these aspects in
the future.
I feel very fortunate to have had such great teachers, supervisors and
colleagues during the last four years. First, I wish to thank the supervisor of my
thesis, professor Esa Itkonen, for his comments on different parts of my study at
different stages. Esa can also be held ‘responsible’ for my interest in linguistic
typology. His class ‘Introduction to typology’ (spring term 1995) and the examples
from ‘exotic’ languages fascinated me and made it clear that this is what I also want
to do. Esa has constantly helped me in numerous ways during the four years I have
been enrolled as a graduate student at the department of linguistics at the University
of Turku and he is clearly the best and the most important teacher I have had. Also
after graduation I have had the pleasure to attend seminars held by Esa or Anneli
Pajunen which has greatly influenced my thinking about linguistics. Anneli has also
helped me in other ways. Discussions with professor Matti Luukkainen during the
writing process of my MA thesis (German) have also helped me a lot.
I also thank the referees of my work, Bernard Comrie and Peter Austin,
4

heartily for their effort and comments on the draft of my thesis. I am very honoured
to have Bernard and Peter as the referees of my study, since their contribution to the
study of transitivity has been very significant, as can be seen, if one takes a look at
the list of references. In 2001 I had the pleasure to be invited to Max Planck Institute
for Evolutionary Anthropology at Leipzig as a visiting PhD student by Bernard
which also offered me a great a chance to meet the people working at the Institute
(Martin Haspelmath also made a significant contribution to this). Discussions with
the people there gave me a lot of ideas. At this point, I would also like to thank
Claudia Büchel and Julia Cissewski heartily for their kindness during my two
months’ visit at Leipzig. Claudia and Julia were always eager to help me and my
family, even if though I was only visiting the Institute. I first met Peter when he was
giving the class ‘Aboriginal and Torres Strait Islander Languages of Australia’ at
LSA 2001 Institute at Santa Barbara. The amount of knowledge Peter has about
these languages deeply impressed me then and still does. Australian languages are
very well presented in my study because of which I am very happy to have Peter as
a referee for my work.
During the four years I have spent working on the present study I have had the
privilege to make numerous great friends and to meet ‘the faces behind the papers’.
What in the beginning amazed me was the eagerness of ‘big names’ to help students
trying to finish a PhD thesis. The following people (in addition to people already
mentioned) have helped me in some way or another (e.g. by giving examples from
languages they are familiar with or by discussing the notion of transitivity with me)
during the four years and it would be unfair not to thank them (people listed in
alphabetical order): Scott DeLancey, Helma van Den Berg, Mark Donohue, Carol
Genetti, Orin Gensler, Katja Gruzdeva, Martin Haspelmath, Timo Haukioja, Soon-
Mi Hong-Schunka, Tuomas Huumo, Nobufumi Inaba, Andrej Kibrik, Hak-Soo
Kim, Ritsuko Kikusawa, Leena Kolehmainen, Magdolna Kovács, Meri Larjavaara,
Frank Lichtenberk, Matti Miestamo, Åshild Naess, David Peterson, Michaela Pörn,
Nick Reid, Sally Rice, Jeanette Sakel, Tiina Savolainen, Eva Schultze-Berndt,
Pirkko Suihkonen, Bertil Tikkanen, Pilar Valenzuela, Viveka Velupillai, Jennifer
van Vorst, Foong Ha Yap and Jussi Ylikoski. All the faults in the study are naturally
my own responsibility.
I feel very fortunate to have been able to work at the Department of
Linguistics at the University of Turku, where I have had a chance to work together
with great colleagues. Discussions on non-linguistic matters, such as hockey (that
is a passion for me and our assistant Timo Haukioja) and fatherhood (Nobu is a
father of three children) have offered me a great opportunity to detach myself from
linguistics on a daily basis. The latter has also been made possible by numerous
‘non-linguistic friends’ with whom it is impossible to discuss linguistics (without
boring them to death), which can be a blessing from time to time. This has
prevented me from taking linguistics too seriously.
Last, but certainly not least, I wish to thank my wife Anna for her support
which has made it possible for me to fully concentrate on my study during the last
four years. The present study is dedicated to my son Jaakko Matias Kittilä, who has
brought nothing but happiness into my life and who has forced me to put things into
perspective and realize that there is a lot more to life than linguistics. The alert
reader will soon notice this if s/he takes a look at the examples from the language
my son is acquiring.
For the funding of my study I thank Emil Öhmannin Säätiö (Emil Öhmann
foundation), Graduate School Langnet, Emil Aaltosen Säätiö (Emil Aaltonen
foundation) and the Linguistic Society of America.

Turku, October 24, 2002,

Seppo Kittilä
6

CONTENTS
NOT APPLICABLE

1. Introduction 11
1.1. Background and aims 11
1.2. Approach and methodology 14
1.3. The structure of the study 19

2. On defining transitivity 20
2.1. Semantic definitions (‘traditional transitivity’) 20
2.2. Structural definitions 21
2.3. Transitivity in discourse (pragmatic definitions) 25
2.4. ‘Multiple definitions’ 26
2.5. Final remarks 29

3. W hat is transitivity (here) 31

3.1. Some relevant notions 31
3.1.1. Ontological vs. ‘conceptual’ transitivity features 31
3.1.2. Transitive valence vs. semantic transitivity of events 34
3.1.3. Interpreted vs. structural transitivity 37
3.2. On semantic transitivity 38
3.2.1. On relevant criteria 38
3.2.2. On the nature of transitive events 32
3.3. Aspects of structural transitivity 46
3.4. Transitivity of meaning and form 55

4. M arking of (high) transitivity 61

4.1. Preliminaries 61
4.2. Linguistic vs. semantic marking 64
4.3. Direct vs. indirect marking 67
4.4. Primary vs. secondary marking of transitivity 72
4.5. On the basic transitive clause 74
4.5.1. Defining the basic transitive clause 74
4.5.2. On the underlying motivation of transitive marking 77
4.5.2.1. ‘Participant’, ‘core-participant’ and ‘agent-patient’ marking languages
78
4.5.2.2. Marking of semantic roles 80
4.5.2.3. Marking of grammatical relations (‘syntactically conditioned
transitivity marking’) 83
4.5.2.4. Reflection of transitivity 84
4.6. A structural typology of the basic transitive clause 85
4.6.1. Theoretical background 85
4.6.2. Verb- and argument marking languages 91
4.6.3. The typology 93
4.6.3.1. Type 1 93
 4.6.3.2. Type 2 96
4.6.3.3. Type 3 99
4.6.3.4. Type 4 107
4.6.4. Summary and discussion 108

5. A typology of transitivity alternations 115

5.1. Transitivity alternation defined 116
5.2. Structural/functional aspects of transitivity alternations 126
5.2.1. Semantically vs. structurally conditioned alternations 126
5.2.2. Lexical vs. morphosyntactic alternations 134
5.2.3. General vs. specific transitivity alternations 138
5.2.3.1. General alternations 139
5.2.3.2. Specific alternations 147
5.2.3.3. Concluding remarks 155
5.2.4. Transitivity increasing, transitivity decreasing and transitivity
rearranging alternations 158
5.2.5. Alternations affecting the number of participants, number of arguments
and individual transitivity features 173
5.2.6. A structural typology of transitivity alternations 188
5.3. Semantic-functional typology of transitivity alternations
215
5.3.1. Direct transitivity alternations 216
5.3.1.1. Minus-transitivizing alternations 216
5.3.1.1.1. Intransitivizing alternations 216
5.3.1.1.1.1. Alternations affecting the number of participants 217
5.3.1.1.1.2. Alternations affecting certain aspects of semantic
transitivity 222
5.3.1.1.2. De-transitivizing alternations 248
5.3.1.2. Transitivizing alternations 254
5.3.1.2.1. Alternations increasing the number of participants (‘genuine
transitivizers’) 255
5.3.1.2.2. Alternations increasing the number of arguments (‘syntactic
transitivizers’) 269
5.3.1.2.3. Alternations increasing individual aspects of semantic
transitivity 271
5.3.2. Indirect alternations 275
5.3.3. Structurally determined alternations 283
5.3.4. Final remarks 288

6. Summary 293

References 303
8

ABBREVIATIONS

A Transitive agent/actor CL Classifier/class

ABL Ablative marker
ABS Absolutive CMPL Completive
ACC Accusative COL Column
ACCID Accidental passive COM Comitative
ACT Active CON(V) Converb
AD Agent demotion CONJ Conjunction
ADD BEN Additional benefactive CONT Continuative
ADEL Adelative CONTESS Contessive
ADESS Adessive COP Copula
ADVERS Adversative CRS Current relevant state
AFF Affirmative CSC Complex stative
AFS Stem affix construction
AFV Verb affix CST Construct state
AFX Affix D/DAT Dative
AGR Agreement marker DC De-control
AGT Agent (marker) DECL Declarative
ALL Allative DEF Definite
ANIM Animate DEIC Deictic particle
ANT Anterior DEM Demotion
ANTIC Anticausative DEPAT De-patientive
ANTIP/AP Antipassive DES Desiderative
AOC Aorist converb DET Determiner
AOR Aorist DETR De-transitive
APPL Applicative DIM Diminutive
ART Article DIR Directional
AS Aseverative DIST Distributive
ASP Aspect DISTPST Distant past
ATTR Attributive DITR Ditransitive
AUG Augmentative DL/DU Dual
AUX Auxiliary DO Direct object
AV Agentive voice DS Different subject
BEN Benefactive DUR Durative
CAUS Causative DYN Dynamic
CFP Clause final particle E Epenthetic schwa
 9

EFF Effective LOC Locative

EL Elative M/MASC Masculine
EMP Emphatic MAL Malefactive
ERG Ergative MASS Generic Patient
EXCL Exclusive marker
F/FEM Feminine MID Middle voice
FACT Factitive MOD Modalis case
FIN Finite MOOD Mood marker
FGR Falling tone grade MS Masculine singular
FORM Formative N/NEUTR Neutral
FREQ Frequentative NAR Narrative case
FS Feminine singular (‘ergative’)
FUT Future tense NEAR Near past
GEN Genitive NEG Negation
GENR Generic NF Non-feminine
HAB Habitual NFUT Non-future
HOR Hortative NML Nominalisation
ILL Illative case NOM Nominative
IMP Imperative NONP Non-past
IM.PAST Immediate past N.PAST Non-past affirmative
IMPF Imperfective NR.PAST Near past
IN/INVOL Involitional NSG Non-singular
INAN Inanimate O Transitive Patient
INC Incompletive OBJ Object
INCH Inchoative OBL Oblique
IND Indicative mood OP Object promoting
INDEF Indefinite affix
INF Infinit(iv)e P/P Past/present
INFR Inferred PART Partitive
INSTR Instrumental PARTIC Participle
INTR Intransitive PASS Passive
INV Inverse PAST Past tense
IRR Irrealis PAST.WIT Witnessed past
ITER Iterative aspect PAT Patient (marker)
ITIVE Itive aspect PERF/PFV Perfective
LD Locative-directional PERI Peripheral participant
LGR Level pitch grade valence increase
LINK Linking particle PL Plural
PM Person marker RM.PAST Remote past
PN Personal name S Intransitive subject
POSS Possessive SA Singular associative
POST Postposition SAF Stem affix
POSTEL Postelative SAG Subject/Agent marker
POT Potential mood SG Singular
PR Pluraility of relations SS Same subject
PRED Predicative STAT Stative
PREF Prefix SUB.ABL Subablative
PREP Preposition SUBJ Subject
PRES Present tense SUF Suffix
PRET Preterite TD Transitivity decrease
PRIOR Prioritive
PRO Pronoun TH Theme suffix
PROG Progressive TI Transitivity indicator
PTCL Particle TMdys Past tense marker, 1
PTV Privative day to 1 year ago
PURP Purposive TOP Topic (marker)
PVB Preverb TR(ANS) Transitive
REAL Realis TRANSL Translative
RECIP Reciprocal TRNZ Transitivizer
REC.PAST Recent past U Undergoer
REDUP Reduplication UNM Unmarked
REFL Reflexive VIS Visual
REL Relativization marker VN Verbal noun
RELINQ Relinquitive VOL Volitional
REQ Requestive X Complex marker on verbs
RES Resultative
1. INTRODUCTION

1.1. Background and aims

Transitivity is a linguistic phenomenon that through the ages has awoken the interest
of linguists. There are numerous studies devoted to some aspect of transitivity. As
early as in Ancient Greece (cf. 2.1.), scholars like Apollonios Dyscolos or Aristotle
developed ideas on transitivity that have been repeated over and over again, yet in
somewhat different forms, in studies also in our day. The basic idea has remained
more or less unchanged, but the growing interest in languages radically different
from Indo-European ones have made further studies of the notion relevant and also
necessary. In addition to the discovery of new languages with exotic structures, the
notion of transitivity has been ‘discovered anew’, which has resulted in a variety of
different approaches to the phenomenon. The basic semantic idea has been rejected
or it has faded to the background in order to let new, more structurally or
pragmatically oriented approaches to the same basic topic flourish. The rise of these
new approaches has also contributed to making the notion even more multilayered
than it was. Consequently, despite intensive study scholars working on the topic still
cannot agree on what actually is transitivity and what would be the best way to
describe and study the notion. In light of studies carried out, this does not surprise
us: transitivity is a vast phenomenon that comprises numerous different facets and
it is extremely difficult to develop a definition that would take account of all this.
This also makes it understandable that transitivity still is a topic of many studies and
will be one also in the future.
There are numerous individual studies concerned with individual features of
transitivity. Just to mention a few: Siewierska 1984, Shibatani 1985 and (ed.) 1988
for passives, Klaiman 1991 for grammatical voice in general, Tsunoda 1988 and
Cooreman 1994 for antipassives, Shibatani (ed.) 1976, Comrie & Polinsky (eds.)
1993, Song 1996 for causatives, Haspelmath 1987 and 1993 for anticausatives,
Kemmer 1993 for middle voice and Frajzyngier & Curl (eds.) 2000 and 2000b for
reflexives and reciprocals. Dixon & Aikhenvald 1997 and (eds.) 2000 illustrate a
more general approach to the whole phenomenon of transitivity. Consequently, due
to the vastness of the topic under study, it is not the goal of the present study to
solve the problem once and for all by discussing every possible transitivity
 13

alternation exhaustively and by proposing a definition that makes further studies

worthless. Since many individual aspects of transitivity have been topics of
numerous detailed studies, it would be perverse even to think that this would be
possible in one study. Hence, the discussion is at some points rather cursory. This
simply must be the case in order to avoid any unnecessary lengthening of our study.
It is, however, our goal to take account of as many features as possible and present
the results in one study. The goal of this is twofold. First, we would like to give an
overview of all the features that can be considered relevant for the notion of
transitivity in general. We wish to show what features of transitivity are relevant and
should be taken account of in studies of individual languages (i.e. what aspects may
have direct effect on the linguistic expression of transitivity). Second, we hope that
the present study will provide the reader an easy access to as many relevant aspects
of transitivity as possible.
The goal of our study was to answer the question: What is transitivity, what
determines the marking and how is the notion expressed linguistically? Originally,
our study focussed on the typology of passives. After this, the goal was to study the
expression of transitivity in German from a typological perspective, but the vastness
of the study finally forced us to drop this aspect as well. We were faced with the
problem that transitivity is a vast phenomenon and that everyone seems to define the
notion somewhat differently (cf. above). As noted above, there are numerous studies
focussing on a certain aspect of this wide and highly interesting concept. What was
lacking was a thorough study concerned with transitivity at a rather general level
trying to define the notion cross-linguistically. We do not wish to ignore the
importance of studies focussing on single languages (see e.g. Dixon & Aikhenvald
(eds.) 2000) from which we have greatly benefited, but a more general approach to
the topic can certainly reveal aspects that have been ignored in studies of individual
languages. Thus, despite making ourselves familiar with many different facets of
transitivity, the present stage seemed somehow unsatisfactory and we felt the need
for a study that could answer the questions above. Perhaps, the best study of the
topic was the seminal paper by Hopper and Thompson that has greatly influenced
our understanding of transitivity as well. Hopper and Thompson (1980) have made
a very significant contribution in that transitivity can be viewed as a continuum (cf.
14

also Duchateau 1998:124, 128) instead of a binary dichotomy.1 The goal of Hopper
and Thompson was to show how certain semantic (defined rather loosely) features
are realized at the level of morphosyntax. They enumerate ten parameters (cf.
below) and show that there are languages in which a given parameter contributes to
the expression of transitivity. In principle, the approach in the present study is the
same. However, there are some things that seem problematic to us. First, as noted
in Tsunoda 1985, some of parameters listed by Hopper and Thompson always co-
vary. Examples include agency and volitionality, the latter is a subpart of the former.
These parameters should not be seen as distinct, but rather as different aspects of an
individual feature. Furthermore, our preliminary study of transitivity had revealed
new aspects worth considering, which also contributed to choosing a comprehensive
typology of transitivity as the topic of our study. One such aspect is illustrated by
changes that are here labelled as structurally motivated (see 5.3.3.). Since our goal
was to illustrate the features that contribute to linguistic transitivity, this cannot be
ignored. Furthermore, a morphosyntactic typology of the alternations, as well as that
of the basic transitive clause2 are completely ignored by Hopper and Thompson (as
well as by other scholars). A detailed study of transitivity alternations requires that
we also take account of the basic structure, since construction A can be justly
labelled as an alternation, if we know what it is derived from. It has proven to be
extremely difficult to define the concept ‘basic transitive clause’ cross-linguistically
(cf. Kittilä 2002), which makes it understandable that the typology of this
construction has been ignored in earlier studies. We were also forced to content
ourselves with a rather simple definition that, however, suffices for our purposes
(see. 5.1.).
What also troubled us in the previous studies of transitivity was the
predominance of an implicit structural approach (cf. e.g. Lyons 1968:350 who

1
The continuum per se is not our concern, since our approach is based on comparison of typical
events with others

2
The term ‘clause’ is used throughout the present study to cover both ‘clause’ and ‘sentence’, as they
are usually used. This distinction is not relevant for our purposes. Moreover, we also focus on the
notion of single clauses and. complex sentences will not be studied in any detail. This is not to say
that the given distinction may not be relevant in numerous other studies.
 15

criticises the traditional use of the term based on semantic intransitivity of clauses
like I hear you). This is mainly reflected in the fact that in many studies only
constructions with a distinct ‘non-active’ verb morphology were labelled as
transitivity alternations. This is, for example, the usual approach to passives (see e.g.
Siewierska 1984:2, Shibatani 1985:837). To us, this kind of restrictive, structurally
dominated, approach seemed unjust in many cases. It is, naturally, true that
languages diverge vastly in what kinds of semantic changes result in a genuine
transitivity alternation and we should not be forcing ourselves to find alternations
where they simply do not exist. However, cases in which the function expressed is
the same and is clearly related to the expression of transitivity, but the structures
employed differ somewhat, should not be ignored. Consequently, we have defined
the concept of transitivity alternations less strictly. Another manifestation of the
implicit structural approach arises in the equal treatment of semantically and
structurally intransitivizing alternations (the latter will be referred as de-
transitivizing alternations in the present study). This generalization is blind to the
underlying motivation of alternations. Without taking account of this, some
alternations may seem arbitrary. In the cases of genuine intransitivization (e.g. he
broke the vase vs. the vase broke), the decrease in transitivity coincides with a
semantic change, whereas de-transitivizing alternations only reduce the structural
transitivity (e.g. the vase was broken). Since the goal of our study is to illustrate
what aspects can result in a transitivity alternation, it is of the utmost importance to
explicitly distinguish between these two alternation types (these terms are used in
many different ways, but in the present study the semantic vs. non-semantic
motivation conditions their use). This only goes for the function, structurally these
alternation types are often expressed by employing the same mechanisms.
Consequently, the ‘usual approach’ is easily understood in this respect.
The basic idea of transitivity in the present study coincides with the traditional
principle that transitive events involve two participants one of which is an agent,
while the other can be labelled as a typical patient. Involvement of two distinct
participants is regarded as the most important transitivity feature here (cf. Croft
(1990:134) who states that no feature is a necessary characteristic of transitivity, but
each feature contributes to the transitivity of clauses). Lack of this makes other
16

features irrelevant. This means that differently from Hopper and Thompson, we do
not claim that (semantically) intransitive clauses could outrank those denoting
events involving two participants regardless of the number of individual transitivity
features. Only events involving two participants have been considered transitive.
The claim of Hopper and Thompson is justified in light of their approach, since
every feature is regarded as equally important and the mere number of features
available is significant. Transitive events necessarily involve an agent and a patient
(defined loosely) and, in our opinion, it is not meaningful to compare events that
involve only one of the relevant features with each other. The fact that the event ‘he
runs’3 has more features of high transitivity than ‘he dies’ does not make it more
transitive (see also (348)-(351) and the discussion). The basic idea noted is taken for
granted also elsewhere and it is not our goal to militate against this fundamental
notion. This would be like trying to prove that two plus two is three. These kinds of
events are the most likely ones to be encoded by transitive clauses cross-
linguistically. Rather, we were concerned with features that motivate the use of
other kinds of structure. This also enabled us to understand the basic concept of
prototypical (high) transitivity better, since by studying the aspects that result in a
deviation from the basic scheme we can learn a lot about what is important for the
given notion. If we are able to show that a given feature results in a transitivity
alternation in language A, this aspect can be considered an important characteristic
of transitivity in that particular language. Furthermore, we may benefit from this
finding in subsequent studies. By studying the typology of these features, we were
able to gain a good overview of what is generally considered relevant for the
expression of transitivity. Even if not expressed explicitly, basic transitive usually
have the given feature. This means that the agency related to a particular event (and
the corresponding clause) is high in case decreased agency produces a transitivity
alternation. Only deviations are explicitly marked in many cases. Furthermore,
agency and affectedness parameters are divided into several subparts in the present
study in order to give a more detailed analysis of these important transitivity
features. It is also in order to note that we were not only concerned with individual

3
The use of single brackets (‘x’) refers to semantics, whereas italics (e.g. he washed me) refer to
actual linguistic data from a language.
 17

transitivity features, but we also the underlying motivation of transitivity is

discussed from different perspectives. These include ontological, semantic and
structural motivations.
As noted above, our goal was to study as many facets of transitivity as
possible. Hence, our goal is not to propose something radically new on the next
pages that would revolutionize the notion of transitivity as we know it today. Many
of the phenomena discussed have been topics of previous detailed studies, but to our
knowledge no individual studies exist that would take account of as many aspects
as have been taken into account here. In a way, our study brings together many
interesting ideas discussed earlier by other scholars. Due to the focus on a particular
aspect of transitivity, a general view has been missing and we aim at filling this
void. A wider scope also aids us in understanding the general notion better. The
broader scope has the consequence that some facets have only been touched upon,
and we hope that someone will find some of the ideas worth studying further.
As noted above, our idea on transitivity has been largely influenced by ideas
presented by Hopper and Thompson in their 1980 paper. There are points on which
we disagree, but it is only fair to say that this paper is the single most important
paper as regards our understanding of transitivity. Later our ideas have been
influenced by many other scholars as well, which has naturally considerably
widened our scope on transitivity. Papers by linguists like (in alphabetical order)
Peter Austin, Barry Blake, Bernard Comrie, Scott DeLancey, Robert Dixon, Mark
Donohue, Werner Drossard, Tom Givón, Martin Haspelmath, Esa Itkonen, Suzanne
Kemmer, Alexandr Kibrik, Andrej Kibrik, Gilbert Lazard, Frank Lichtenberk,
Marianne Mithun, Anneli Pajunen, Masayoshi Shibatani, Sally Rice, Jae Jung Song
and Helma Van Den Berg have also made me understand the notion of transitivity
a lot better. In general, we have to say that the ideas presented in the subsequent
chapters illustrate a kind of combination of ideas presented by others. Ideas of others
have forced us to modify our ideas, which is most fruitful for one’s thinking. It is
simply impossible to name all the people, whose contribution can be considered
relevant. I apologize, if I unintentionally present ideas of others as our own and
forget to cite the people behind the ideas (this should not be the case at any point of
our study).
18

1.2. Approach and methodology

As noted above, the primary goal of our study is to define the concept of transitivity
as exhaustively as possible (yet at a rather general level) by showing what semantic
and structural aspects can result in a transitivity alternation. The notion of
transitivity alternations is used in a somewhat unorthodox way on the next pages
(see section 5.1. for a detailed definition). Every possible change in the transitivity
of clauses illustrate a possible transitivity alternation. This means that not only
passives, antipassives, causatives etc. are taken account of, but also cases in which
the structure of the basic transitive clauses is only mildly affected have been
included. The latter comprises, for example, differential object marking that is
usually not labelled as a transitivity alternation. Prerequisite for an alternation to be
labelled as a transitivity alternation is that it is somehow motivated by changes in
semantic transitivity or that the result is similar to such an alternation. In general,
we may perhaps say that we illustrate the interaction of syntax and semantics in light
of transitivity. The primary starting point in this is semantics. However, since a
semantic aspect can be considered relevant for the notion of transitivity only if it
coincides with a structural change, we cannot but observe the structure first in many
cases. Furthermore, we were also forced to study most languages through a language
we know and rely on the translations. This posed any problems extremely rarely, if
at all. The primacy of semantics refers to the fact that we always try to explain the
structural phenomena by referring to semantic transitivity. If there is not a valid
explanation to be found, the structure in question has been ignored (cf., however,
5.3.3.). Furthermore, semantic evidence obtained from other languages can be used
as a starting point when working on new languages. The primarily semantic starting
point enables one to de-emphasize structural differences due to the nature of
languages. This means, for example, that both accusative and ergative basic clauses
can be regarded as equally transitive based on their semantics.
The approach adopted in the present study is purely synchronic. The
diachronic development that has resulted in the alternations represented has not been
taken account of in any way, even if this might have aided us in gaining more
thorough insights into certain phenomena. The typological variety of alternations
 19

can likely be traced back to different origins of the given alternations. This is also
ignored in many grammars of ‘exotic languages’ and second, adding diachrony to
the present study would have made it even more bulkier.
Before proceeding to details, it is of the utmost importance to define what we
understand under the concepts in question. Transitivity is here regarded primarily
as the linguistic realization of non-linguistic transitivity. Non-linguistic transitivity,
on the other hand, comprises aspects that have to do with the nature of participants
and events along with the efficiency of the energy flow (a more detailed definition
will follow in chapter 3). The basic transitivity is based on the presence of distinct
agent and patient participants, whereas every possible change in the structure of
clauses, motivated by some feature of transitivity, is regarded as a possible
transitivity alternation. In typical cases, the change is reflected directly through
morphosyntactic changes in the clause structure, but there are also cases in which
the expression is less direct. This kind of approach to the notion of transitivity
alternation suffices for our purposes, even if it is far from being without problems.
The main problem we might have to face is that the obvious danger of circularity.
In order to avoid this, we tried to use semantic evidence to justify our claims
whenever possible. Put concretely, this means that we always tried to validate our
claims by referring to the structure of clauses denoting semantically less transitive
events, i.e. we use non-linguistic evidence when arguing for linguistic facts. In ideal
cases, we were able use data from the very languages in question, but in some cases
we were dependent on cross-linguistic data, which is clearly less reliable in this
respect. There are also cases in which we can label a change in the clause structure
as a transitivity alternation primarily by referring to structure.
The primary focus lies here on the morphosyntactic realization of semantic
transitivity, which is studied employing the approach outlined above (for a more
detailed analysis, see 5.1.). Semantic transitivity (in the sense the notion is
understood here) comprises two different notions both of which make a significant
contribution to transitivity. These have been labelled as inherent and contextual
transitivity. The former refers to differences between events like ‘he killed the bear’
vs ‘he saw the bear’, whereas the latter comprises cases in which the inherent
transitivity of events is somehow manipulated. This means, for example, that the
20

degrees of agency may vary. Changes in inherent transitivity are motivated by

changes between fully distinct events, while changes in contextual transitivity do not
affect the basic semantics of events. In our study, the latter is focussed on. This
enables one to focus on individual features of transitivity a lot better. This is very
important, since our goal is to illustrate the underlying motivation as exhaustively
as possible. Agency and affectedness vary independently on each other, which
makes it necessary to use an approach that enables one to take account of these
differences. If an alternation is caused by changes in inherent transitivity, more than
one feature is usually affected and it might be impossible to analyze the underlying
motivation more closely. Furthermore, the semantic and structural variety of
alternations due to changes in contextual transitivity is far greater than that related
to changes in inherent transitivity only. Effects of inherent transitivity have not been
ignored, either, but only discussed in less detail. The conditioning factor is clearly
different from that of contextually induced alternations and ignoring this aspect
altogether would result in ignoring some relevant features.
Not only the underlying motivation of the alternations, but also the structural
typology of alternations is of the utmost importance for our study. In general,
different alternations differ from each other in how obvious their relation to genuine
expression of transitivity is. A generalization holds that alternations expressed by
manipulating the number or case marking of arguments are more obvious than those
expressed solely by modifying the verb morphology (in the latter case changes in the
former aspects are excluded). This is due to the fact changes in the case marking are
rather often motivated by semantic changes (cf. above). In addition to discussing the
semantic transitivity in detail, we also propose a structural typology of basic
transitive clauses and transitivity alternations. Semantic and structural aspects are
largely discussed independently of each other (cf. 1.3.). The expression of basic
transitivity and transitivity alternations differ crucially from each other in this
respect. In the former case, semantic and structural transitivity coincide with each
other, i.e. a clause that denotes a highly transitive event is regarded as the basic
transitive clause of a given language.4 This is a methodological necessity, since in

4
The picture is a lot messier than this, since semantics is not alone responsible for the structure of
clauses.
 21

order to be able to study transitivity alternations, we have to be able to define the

notion of high linguistic transitivity as unambiguously as possible. Semantic
transitivity is clearly the most appropriate starting point in this respect, since, as will
be shown (and as is generally known), the marking of highly and less transitive
events can be radically different. All of the above does not mean, however, that we
would restrict the concept ‘basic transitive clause’ to a specific kind of structure
(e.g. to a typical nominative-accusative construction). In the case of transitivity
alternations, we will not aim at proposing clear correlations between form and
nature of alternations. This means that a certain structural alternation can be
employed in the expression of numerous different semantic alternations. Some
generalizations can be drawn (e.g. Patient omission is typically conditioned by
indefiniteness), but these are too general to have any relevance for the present study.
The rather general approach adopted also implies that our goal is not to enable
a typological classification of languages. Rather, our goal is to illustrate the
functional variety of transitivity alternations, i.e. the emphasis is not on any
particular alternation, which should be the case, if we aimed at proposing a tool for
classifying languages. Also here, the discussions of basic transitivity and transitivity
alternations differ crucially from each other. A language can have only one clause
type that can be regarded as the basic transitive clause5. On the other hand, every
language is capable of expressing a variety of transitivity alternations that display
obvious interlingual differences. This means, for example, that language A has a
transitive reflexive, a morphologically distinct passive construction, but has no
middle voice, whereas language B expresses the two former alternations by a
general intransitivizing affix without having any kind of middle voice construction.
The structure of particular transitivity alternations in a given language is not
inferable from the structure of others. There are no significant generalizations to be
drawn, which means that it is impossible to classify languages according to this
feature.6 This does not mean, however, that the notion of typological classification

5
Even if it might be extremely difficult in some cases to definitely state which of many possible
structures is accorded the status of basic transitive clause, see e.g. Toratán (Himmelmann & W olff
1999) and Philippine and Formosan languages in general (Peter Austin, p.c.).

6
W e hope that future studies will change this.
22

is completely irrelevant for transitivity alternations. Individual transitivity

alternations like passive, reflexive, antipassive etc. are naturally dividable into
different structural types. As regards the typological classification of languages on
the basis of the structure of individual transitivity alternations, the best we can do
is to label languages as having agentless passives, transitive reflexives etc., but this
is very unlikely to correlate with any other structural facet, and cannot thus classify
languages. A detailed structural typology of each functional transitivity alternation
type would have made the study unnecessarily long, because of which this aspect
has been ignored. We do, however, present different examples of each type for the
purpose of illustrating the structural variety of individual alternations.
As is typical of linguistic typology, studying grammars of individual languages
was the chief method of our research. It is in order to thank the authors of grammars
used in our study that would not have been possible without the efforts of other
linguists (people to be thanked find their names in the references). Prior to
ploughing through numerous grammars we made ourselves familiar with individual,
more theoretically oriented, studies of transitivity. In addition to enhancing our
knowledge on the topic, these have also been valuable sources for illustrative
examples. These two independent, yet closely related approaches to our topic, have
made it possible for us to understand the notion of transitivity rather broadly.
Individual grammars have provided us with the possibility to present some of our
ideas concretely by illustrating the concept in question by referring to concrete
examples. On the other hand, individual grammars have also made modifications
of some ideas necessary, since we have been faced with aspects not taken account
of earlier. As regards the structural typologies proposed, individual grammars have
made the most significant contribution to our study. The initial focussing on
theoretical basis of transitivity enabled us to focus on what is relevant for our
purposes. This has been of the utmost importance for the structural typology of
different alternations. We were to focus on collecting examples of cases that can
definitely be regarded as transitivity alternations. Since individual grammars are
general descriptions of whole languages, transitivity is only one of the aspects
discussed in them. In some cases, this naturally has the consequence that some
interesting facets are ignored in a given grammar. Other grammars, on the other
 23

hand, are more concerned with the notion of transitivity. On account of these
differences, we have had to work on a ‘take-what-is-interesting principle’, which
means that grammars (or other relevant studies) in which transitivity is thoroughly
discussed, dominate as regards the selection of examples. This simply could not be
avoided. This has the consequence that at some points interesting phenomena may
be illustrated in light of examples from a single language, which does not mean that
the given feature could not be relevant in other languages as well. Another problem
related to selecting examples follows from the obligatorily chronological process of
checking grammars for examples. In some cases, an interesting phenomenon has
presented itself too late, because of which it has naturally been ignored up to the
point of its discovery. The latter problem is not very severe though, since the
ignorance of a given aspect is in many cases due to the fact that the phenomenon in
question has not been discussed or is not of great significance.
A further interesting ‘problem’ related to the selection of examples is
presented by the obvious dominance of predominantly ergative languages in our
data. The percentage of ergative languages used in the illustration clearly outranks
the percentage of ergative languages in general (that according to Dixon (1994:2)
is about one quarter, i.e. 25 per cent). We cannot give any valid explanation for this
unequal distribution of language types. It simply seems that ergative languages are
for some reason (yet to be discovered) more sensitive to explicit expression of
transitivity. One possible explanation might be found in the importance of agency
for high transitivity in some ergative languages due to an explicit marking of the
Agent7 and the agency associated with it (cf. e.g. Wierzbicka 1998:170, for a
different kind of analysis see Comrie 1978:355ff). In order for this to have any
greater significance, the former claim should be validated by a detailed study. A
further (structural) aspect that possibly makes a minor contribution to the
predominance of ergative languages is presented by the predominantly accusative
nature of all analytic languages. Analytic languages lack the means to express a rich
variety of transitivity alternations by manipulating the marking of arguments or verb
morphology, which makes them less interesting for our typology. Another (rather

7
According to Dixon (1987b:3) ergative is always the explicitly marked case in ergative systems, a
possible (yet rather marginal) exception is found in Roviana, see Corston 1996:15).
24

marginal) explanation might be that authors working on ergative languages are more
concerned with transitivity phenomena, because of which the notion is dwelled on
in grammars or studies written by them. This does not seem very plausible, though.
It merits a notion in this context that also in Hopper and Thompson’s 1980 paper the
ratio of ergative languages is approximately 40 per cent, which is also clearly higher
than the ‘expected’ 25 per cent. This possibly provides us with further evidence for
the fact that ergative languages are for some reason more sensitive to transitivity
than accusative ones.
In general, we may say that the present study is rather data-oriented. We have
focussed on typological illustration of the relevant phenomena. Theoretical
discussions are limited to a minimum, which does not mean that this aspect is
disregarded. Many facets that may be regarded as relevant by others will be ignored.
One example of this is provided by the rather brief discussion of causatives. As
noted, for example, by Song (1996:1), causative constructions have been one of the
most recurrent research topics in linguistics. There are numerous studies devoted to
causatives only (Shibatani (ed.) 1976, Comrie and Polinsky 1993 and Song 1996
just to mention but a few). We have, however, ignored some aspects simply in order
to save space. The obvious focussing on data is also reflected in the fact that the
theoretical discussion has been greatly influenced by our typological findings. Due
to the emphasis on data, the present study may also serve as a valuable source for
examples. A reader concerned with the notion of transitivity will find the data
presented by different authors in different contexts and different studies packed in
one.
Before proceeding to the topic itself, it is in order to emphasize that, even if
the present study illustrates our ideas on transitivity as we see the concept now,
future studies may make revisions necessary. This inheres in the general approach
adopted. Detailed studies of certain aspects may reveal something that has been
ignored by us. This is not a problem, however, since, as noted above, we do not aim
at explaining every aspect of transitivity exhaustively, but our goal is to show what
features are in general relevant for the concept of transitivity.
 25

1.3. The structure of the study

The present study comprises three independent, yet closely related main parts. The
first part discusses the theoretical basis of transitivity. We begin by briefly
examining different ways in which transitivity has been defined in during the history
of linguistics. This presentation is not very detailed and only a couple of illustrative
definitions of each type are taken account of. After this, we will present some
relevant facets associated with the notion from our own point-of-view. At this point,
the number of concrete examples is limited to a minimum. It is in order to say that
the theoretical discussion has at some points been greatly influenced by our findings
(cf. above). In the second part, the notion of high linguistic transitivity will be
discussed. The notion is based on the linguistic expression of basic transitive events.
We will discuss the rationale behind the marking and propose a structural typology
of the basic transitive clause. The goal is to show what structural features can be
considered relevant for the linguistic expression of high transitivity. In the last part,
we will dwell on the notion of transitivity alternations. This part comprises two
subparts. We begin by presenting a structural typology of transitivity alternations.
At this point, the underlying functions of alternation are irrelevant. In the second
part of the last chapter, we will illustrate the underlying motivations of alternations
as exhaustively as possible. On the basis of the motivation and the structural
realization, three major types will be distinguished. These are labelled as direct,
indirect and structurally motivated alternations.
2. ON DEFINING TRANSITIVITY

The notion of (linguistic) transitivity was originally introduced in Ancient Greek by

scholars like Apollonios Dyscolos and Aristotle (see, for example Baratin 1998:
16ff, Samain 1998:39). Since then, transitivity as a linguistic phenomenon has been
defined in a number of different ways and it is in order to discuss some definitions
in more detail before proceeding to proposing a typology of the phenomenon.
Consequently, in the present chapter examples of semantic, structural and pragmatic
transitivity definitions are illustrated and briefly discussed. Different transitivity
features will be discussed separately in light of different definitions that emphasize
the aspects in question. Presented definitions are explicitly distinguished from each
other despite obvious overlaps in many cases (see, for example, the list of
transitivity features outlined by Hopper and Thompson in 1980 (p. 252)). The goal
of this chapter is not to dwell on different transitivity definitions, but only to give
a brief overview of the relevant features. Only a few typical instances of each type
are illustrated for the purpose of exemplifying the variety attested. Individual
transitivity features (that are relevant for our purposes) will be discussed thoroughly
in the following chapters.

2.1. Semantic definitions (‘traditional transitivity’)

The traditional view on transitivity suggests that the effects of an action pass over
from agent to patient (cf. e.g. Lyons 1968:350 and Tsunoda 1994:4671 and below).
This coincides largely with what is here considered as semantic transitivity. A
definition is best considered semantic, if it emphasizes facets that can be justly
viewed as semantic (or non-linguistic). These include aspects like affectedness of
the patient and agency (see e.g. Hopper and Thompson 1980:252). Structural
properties of clauses as such are rather irrelevant. This means, for example, that
accusative and ergative constructions are both regarded as typical transitive
constructions despite their clear structural differences. Highly transitive features of
clauses are based on the semantic transitivity of profiled events. The most traditional
definitions were semantic in nature.8 Traditionally, (linguistic) transitivity was

8
Even if the use of the notion differs from that illustrated here, e.g. structure was also important.
 27

understood as an efficient energy flow from one participant (agent) to another

(patient) (cf. e.g. Lazard 1998b:55).9 Constructions that describe these kinds of
event were considered transitive, while others were not. Semantic definitions are,
of course, not restricted to the classical period, but many modern definitions have
also adopted a semantic approach. The basic idea of these definitions is the same.
Typical examples of (modern) semantic transitivity definitions are, for example, the
following:

In accordance with the etymology of the term, semantic transitivity evokes the idea of
something passing (transit) from one participant to the other, from agent to the object.
W e are led to think that a sentence meaning, for instance “the gardener killed the
rabbit” is typically transitive, since it implies some intention in the agent which is
realized in the action, whose effect is to modify the state of the object: from the will
in the gardener’s mind something is passed into the outer world, a thing which is
manifested in the fact that the rabbit is dead. (Lazard 1998:236)

Semantic definition of transitive event

a. Agent: The prototypical transitive clause involves a volitional, controlling, actively-
initiating agent who is responsible for the event, thus its salient cause.
b. Patient: The prototypical transitive event involves a non-volitional, inactive non-
controlling patient who registers the event’s changes-of-state, thus it’s salient effect.
c. Verbal modality: The verb of the prototypical transitive clause codes an event that
is compact (non-durative), bounded (non-lingering), sequential (non-perfect) and
realis (non-hypothetical). The prototype transitive event is thus fast-paced, completed,
real, and perceptually and/or cognitively salient. (Givón 1995: 76)

Event transitivity encodes a model of directed, unilateral “energy transfer” (Langacker

1990:218) from Agent to Patient. The Agent is the source of energy transfer, whereas
the Patient is the participant targeted by it; The Patient is either created, manipulated
or changed in the process. (Davidse 1996:98)

The definitions outlined by Lazard, Givón and Davidse are purely semantic in the
sense that they do not take into account the morphosyntactic structure of clauses
employed in the description of transitive events. Only semantic criteria as, for

9
The terms ‘agent’ and ‘patient’ were introduced much later.
28

example, volitionality of the agent, affectedness of the patient etc. are relevant (for
a more detailed analysis of relevant semantic parameters, see 3.2.). This means that
the kind of event described above can be expressed linguistically in many different
ways all of which must be regarded as transitive due to the semantic nature of the
definitions at issue. This can be viewed as both the strength and the weakness of
these definitions. First, because the morphosyntactic form of clauses is irrelevant,
these definitions are capable of describing transitivity in all languages despite their
structural differences. This means, for example, that accusative and ergative
structures are both considered transitive, since they both describe similar events.
Consequently, semantic definitions are clearly the best definitions as regards the
study of typology of transitivity. Second, however, semantic definitions (in a strict
sense) cannot separate structurally different constructions from each other. This
means that all constructions that can be used to describe semantically transitive
events must be viewed as equally transitive. For example, both active and passive
are transitive constructions according to semantic definitions, since they both
describe semantically transitive events.10 Semantic definitions can consequently be
rather inapplicable to the study of transitivity in a single language, even if their
typological applicability cannot be denied.

2.2. Structural definitions

The originally semantic idea of transitivity has also given rise to many primarily
structural transitivity definitions. This is quite natural, since semantic transitivity
coincides with structural transitivity in typical cases. Structural definitions have
perhaps even become the more significant ones as a result of the rise of formally
oriented theories of grammar. It should be rather safe to say that most of us are more
familiar with structural transitivity definitions than with traditional semantic ones.
For semantic definitions, the starting point in the study of transitivity is the
nature of non-linguistic events. Any structure that denotes a typical transitive event
(such as those described by Lazard, Givón and Davidse above) is considered

10
They can both also describe events whose semantic transitivity is very low, but as long as the event
denoted is the same, both of these clause types must be considered semantically transitive.
 29

transitive. Since most (or even all) of the languages that were targets of description
in the classical period and also later were nominative-accusative Indo-European
languages (Latin, Greek, English) the transitive archetype was any two-actant clause
with a nominative (unmarked) subject and an accusative object. The predominance
of nominative-accusative languages has in some cases resulted in the fact that any
structure that deviatess structurally from typical nominative-accusative structures
is deemed intransitive. This is quite unfortunate, since this approach excludes many
semantically transitive constructions.
One characteristic of structural definitions is that their variety is far greater
than that of semantic ones. Semantic definitions are all based on the same principle,
i.e. on the efficient and unilateral energy flow from one participant to another (i.e.
from agent to patient). All structural definitions have in common that the number
of explicitly expressed arguments must be two, but the definitions diverge in
whether they take other aspects into account as well. Semantic transitivity is
typically ignored in most structurally oriented transitivity definitions. In the purest
cases, only the number of arguments is relevant. An example of this kind of
definition is given below, cf.

Transitivity in natural language is commonly approached in one of the two ways. One
approach, owing its origin to predicate logic, defines transitivity in terms of the
number of noun arguments necessary to make a predicate coherent. A predicate
requiring only one such noun argument is termed intransitive and a predicate requiring
two or more transitive. This definition is blind to the relationship obtaining between
the two arguments, according equal transitive status to English verbs such as differ,
resemble and meet as to the verbs hit and eat. In a language marking case, no
difference in transitive status is accorded to verbs requiring different case patterns, as
long as the number of obligatory nouns is the same. (Jacobsen 1985:89)

As explicitly stated by Jacobsen, the kind of definition outlined here originates from
predicate logic. This results in the very pure structural nature of the definition. As
also pointed out by Jacobsen, the semantics of the clauses is irrelevant as long as the
predicate requires two or more arguments. The form of the expressed arguments is
also ignored in these kinds of definition. Hence, these definitions are purely binary
in nature. Clauses always have one or two (or three) arguments, there are no clauses
30

with one and a half arguments, for example. Consequently, definitions like that
illustrated above make a strict division of clauses into transitive and intransitive
possible. These definitions are best considered syntactic, since they are completely
based on syntactic criteria.
Most structurally oriented transitivity definitions are not as absolute as that
illustrated above. In most cases, not only the number, but also the form of arguments
is taken account of. For example, usually only clauses having a direct (accusative)
Patient or an ergatively marked Agent.11 are considered transitive, while all others
are deemed intransitive despite the number of explicitly expressed arguments. Not
only syntax, but also morphological marking of arguments is important for these
definitions that are illustrated below, cf.

Die Einteilung in transitive und intransitive Verben beruht auf dem Verhältnis des
Verbs zum Akkusativobjekt und der Sonderstellung des Akkusativs unter den
Objekten. Transitive Verben sind solche Verben, bei denen ein Akkusativobjekt stehen
kann, das bei der Passivtransformation zum Subjektsnominativ wird. [...] Intransitive
Verben sind solche Verben, bei denen kein Akkusativobjekt stehen kann, unabhängig
davon, ob ein anderes Kasus- oder Präpositionalobjekt bei ihnen stehen kann. (Helbig
& Buscha 1993:53)

Intransitiv vs. transitiv bedeutet im Bewusstsein der Grammatiker und der Sprecher
gemeinhin, dass - je nach Kodierunstyp - transitive Verben einen Kasusrahmen wie
ERG-ABS oder NOM-AKK fordern, intransitive jedoch nur einen NOM oder ABS.
(Drossard 1991:165)

Of these definitions, the former is used in the description of German (alone),

whereas the latter has more typological relevance. In principle, these definitions are
as binary as that cited by Jacobsen, because they also recognize only two clause
types with no intermediate forms. However, the concept ‘transitive clause’ is here
somewhat more strictly defined and distinguished from less transitive constructions

11
Throughout the present study, initial capitals on Agent and Patient refer to grammatical arguments,
whereas the lack of capitals, i.e. ‘agent’ vs. ‘patient’ refers to semantic roles or participants. in
distinction from the traditional use (see e.g. Palmer 1994:6) ,the labels Agent and Patient are not
restricted to mere grammatical roles, but the label covers here all possible kinds of reference to the
instigators (agents) and targeted participants (patients) regardless of the morphosyntax of arguments.
 31

on the basis of a higher number of relevant criteria. Not any construction with two
explicitly expressed arguments is considered transitive, but a clause must
correspond to a more detailed prototype in order to be regarded as transitive. In this
case the morphological case marking of arguments is also relevant. Consequently,
these definitions could be called morphologically (and syntactically) transitive.
Definitions like those presented by Helbig & Buscha and Drossard emphasize more
the semantically transitive nature of certain kinds of clause that are usually
employed in the description of semantically transitive events in any language.
Hence, the transitive nature of basic accusative and ergative clauses is indirectly
motivated by semantics. However, structural definitions like these ignore some
semantic differences, since clauses like he killed me and he saw me are both
considered transitive despite their semantic differences (for a justification for the
relevance of these differences, see section 5.3.2.).
Some structural definitions ‘go even further’ in taking certain indirectly
observable morphosyntactic properties into account. A typical example of such a
feature is the passiviability of clauses (see also 5.3.2.). In some definitions, only
clauses that allow passivization are transitive, while others are classified as
intransitive. This restriction is also relevant for Helbig & Buscha (see above). In
German (as in many other languages as well), passivization is not a property of all
transitive clauses, even if most transitive clauses may be passivized. This criterion
excludes certain clauses with accusative objects, as well as all other objects (e.g.
dative and genitive objects) from high transitivity. Typically, the excluded clauses
are rather low in their semantic transitivity, i.e. only clauses that are somehow
conceptualized as transitive allow passivization. Consequently, passivization is a
sufficient criterion for high transitivity in many cases. It restricts the notion of the
basic transitive clause even further than do pure morphological criteria.
Passivization is not a property that could be directly inferred from the morphological
marking of clauses, but it can be observed only through a more detailed examination
of clauses.
Passivization makes it in many (but not all) cases possible to separate
transitive clauses from less transitive ones, since, as noted, only clauses conceived
of as somehow transitive are to be passivized in many languages. The acceptability
32

of passivization correlates to some extent with transitivity: the more transitive a

clause is, the more readily it can be passivized (see, for example, Lehmann
1991:224f and Rice 1987). We do not expect to come across a single language that
passivizes constructions low in semantic transitivity, but cannot passivize highly
transitive constructions (according to definitions given in 2.1.). Furthermore, the
criterion of passivization can distinguish between different instances of the same
clause, something that other structural definitions are incapable of. For example, the
clause *the couple next door is known by John is not acceptable in English, whereas
the couple next door is completely known by John is fully grammatical. The
construction per se is the same, the only difference lies in the presence of the adverb
completely in the latter clause. This makes the latter clause semantically somewhat
more transitive, which results in the passiviability of an originally unpassiviable
construction (for a very detailed study of transitivity in English based on the idea of
passivization, see Rice 1987).
Even if passivization is in many cases an adequate transitivity criterion, we
have to be cautious not to ‘overuse’ it. The notion of transitivity cannot be based
solely on it. First and foremostly, there are many languages that lack a genuine
passive (or antipassive). It goes without saying that we cannot use passivization as
any definition for transitivity in these kinds of language. Secondly, there are also
many languages in which also syntactically intransitive clauses allow passivization.
If we rely exclusively on passivization when defining transitivity, many intransitive
clauses have to be viewed as transitive as well in these languages. Third, there are
languages in which other arguments than direct objects may be promoted to the
subject of passive. In these cases, passivization does not distinguish between
transitive and intransitive constructions, either, since the promoted element may be,
for example a locative circumstant of an intransitive clause. Furthermore,
passivization is an applicable criterion only in predominantly accusative languages.
However, the equivalent of passivization in predominantly ergative languages,
namely the antipassivization has also been used as a defining definition for
transitivity, cf.

In sharp contrast, antipassivization can only apply to two-place (or more-place) clauses
with the respective transitive case frame, e.g. ERG-ABS, NOM-ACC and NOM-ABS
 33

in W arrungu. But, it can never apply to one-place clauses or two-place clauses with
a non-transitive case frame, e.g. ABS-DAT and ABS-LOC. (Tsunoda 1988:636).

Since the antipassive is a kind of mirror-image of passive, it is very natural that the
antipassivization reflects to some extent transitivity of clauses as well. This seems
to be the case at least in Warrungu, since only clauses with a transitive case frame
allow antipassivization.12
There are also other problems relating to structural definitions that are due to
the immense linguistic variety of languages. One rather obvious fault is related to
the terminology employed in typical structural definitions. Labels ‘subject’ and
‘direct object’ are very often used in structural transitivity definitions. The whole
notion of transitivity may even be defined solely on the basis of these terms, cf.
(Siewierska 1991: 73):

In traditional grammar, the subject and object are characterized in relation to the
notion of transitivity, which itself receives no independent definition.

The use of subject and object in transitivity definitions follows from the
predominance of accusative Indo-European languages as the basis of the proposed
definitions (cf. above). In Indo-European languages, the notions of subject and
object are rather easily definable. Because of this, structurally oriented transitivity
definitions can be used in the description of transitivity in these languages.
However, subject and object are (as generally known) cross-linguistically very
problematic notions (see e.g. Keenan 1976). Because of their Indo-European origins,
subject and object are burdened with structural properties of IE languages. Most
languages are, however, not like Indo-European ones. If we hold on to these notions
when defining transitivity in structurally ‘exotic’ languages in which subject and
object are marginal at best, we run into difficulties (see the citation by Lazard
below). If the whole notion of transitivity relies on these concepts, transitivity
cannot be defined in languages that lack them. We cannot but classify these

12
Transitivity could also be inferred directly from the employed case frames. W e are not aware of any
study that would have focussed on the relation obtaining between the ability of clauses to form
antipassives and transitivity.
34

languages as inherently intransitive. This kind of classification is not unheard of, but
some predominantly ergative languages have been defined as inherently intransitive,
since they lack a marked direct object. Moreover, the argument that coincides with
the subject of accusative systems is marked (cf. e.g. Job 1985 and Dixon 1994: 22
for comments on Mel’ uk 1983 and Mel’ uk 1992: 104f). Even if the starting point
of a definition is the structure, structurally different languages have to be taken into
account. Subject and object could be replaced, for example, by S, A and O
introduced by Dixon (1979), since these terms are more applicable cross-
linguistically.
Structural definitions differ furthermore from each other in whether they view
transitivity as a property of verbs or clauses. A verb that can have a direct
(accusative) object is considered transitive, whereas all others are deemed
intransitive. An example of a typical verb-based definition that is usually found, for
example, in traditional grammars is cited by Lazard (1998: 160), cf.

For traditional grammar, transitive Verbs are those that take a direct object or an
object in the accusative: such is the construction of action verbs and assimilates; all
the remaining verbs are intransitive. Such a conception is only valid in the case of
accusative languages; besides, it does not deal with differences between constructions
other than the major construction.

Definitions that are primarily based on transitivity of individual verbs have been
(rightly) criticized, because most verbs (in almost any language) can be used
transitively (e.g. he eats meat) or intransitively (he eats, see, among others,
Bausewein 1990: 34). In similar vein, many verbs that are considered primarily
intransitive (e.g. he is singing) can get a cognate object (e.g. he is singing the
national anthem of Finland) and be used transitively. Consequently, the terms
‘transitive’ and ‘intransitive verb’ (in a very strict sense) are quite unfortunate in
many respects. These notions should be used only as umbrella terms for a general
notion of the two kinds of verb. Their use in the classification of individual verbs
causes problems. The latter kind of use is very closely related to defining transitivity
as a property of clauses instead as one of individual verbs. If we adopt this approach,
we can explain the ‘peculiar’ nature of ambivalent (or labile) verbs by referring to
 35

the transitivity of clauses. Transitive and intransitive clauses are easily defined as
clauses with two explicitly expressed arguments (that, depending on the definition,
have to appear in a certain form). There are verbs that can appear in both of them.
These kinds of definition are furthermore better than those based on the transitivity
of individual verbs, because the expression of arguments is only indirectly (even if
perhaps primarily) determined by verbs. Omission or introduction of arguments is
often due to other properties (for example, to the definiteness of arguments).

2.3. Transitivity in discourse (pragmatic definitions)

In addition to the widely accepted semantic and structural definitions, transitivity

has also been defined emphasizing other aspects of transitivity. One such facet is
transitivity in discourse (term adopted from Hopper and Thompson’s 1980 paper).
A definition is here classified as pragmatic (due to a lack of a better umbrella term),
if it is primarily based on other transitivity features than (morpho)syntax or
semantics (or transitivity in the non-linguistic world). According to typical
pragmatic definitions (in the sense the term is used here), transitivity is primarily
conditioned, not by the semantics of events, but by contextual factors that impinge
on the expression and marking of arguments in a very significant way. A rather
typical definition that can be justly labelled as pragmatic, is cited in the following:

W e should constantly remind ourselves that the number of syntactic core arguments
depends not on the number of entities involved in the situation referred to, but on the
manner in which the situation is conceptualized by the speaker, and that one cannot
speak, for example, of a “transitive action” or “intransitive action”, because the same
action may be viewed as “transitive” or “intransitive” depending on the point of view.
(W ierzbicka 1996: 410)

In the view of Wierzbicka, we are not justified to speak of transitive or intransitive

actions, since the number of arguments does not follow primarily from the number
of participants partaking in events. More relevant for the transitivity of clauses is the
way in which speakers conceptualize the event. The view adopted by Wierzbicka
is in this respect very justified, because, as already briefly noted above, the
36

expression of arguments is only indirectly conditioned by semantics. However,

Wierzbicka should have taken account of the primacy of semantics. Even if many
‘non-ontological’ factors make a very significant contribution to the expression (and
marking) of arguments, it is, however, usually the ontology (or semantic transitivity
in more general terms) that makes the expression possible. If the semantics of a
particular verb does not allow its object to be deleted, the number of core arguments
is obligatorily two irrespective of the conceptualization of the event by the speaker.
Similarly, the semantics of many intransitive verbs does not allow any kind of object
to be introduced. These kinds of definition are furthermore blind to the use of
different case frames in the expression of events that differ from each other in
aspects other than transitive valence. For example, in languages in which
experiencer constructions are structurally less transitive than typical transitive
events, the conceptualization (in the sense of Wierzbicka) is irrelevant, since only
highly transitive events can be encoded transitively. This is clearly due to
ontological differences between the given events. As a consequence, definitions like
those proposed by Wierzbicka are in many cases insufficient.
The definition proposed by Wierzbicka is primarily based on the expression
of (core) arguments, the changes in the marking of arguments in different situations
is not taken into account. Pragmatic factors affect, however, also the marking of
arguments in a very significant way. One important aspect of the differential
marking of arguments is the foregrounding/backgrounding distinction noted in their
seminal paper by Hopper and Thompson (1980: 294):

W e have made and supported the claim that Transitivity is a global property of
clauses, that it is a continuum along which various points cluster and tend strongly to
co-occur, and that the foci of high Transitivity and low Transitivity correlate with the
independent discourse notions of foregrounding and backgrounding respectively. The
fact that the semantic characteristics of high Transitivity such as perfective Aspect,
individuated O, and agentive Subject tend strongly to be grammaticalized in the
morphosyntax of natural languages points to the importance of the
foregrounding/backgrounding distinction [...].

According to Hopper and Thompson, foregrounding/backgrounding distinction is

the most important single feature of transitivity, since many semantic facets
 37

correlate with it. The definition illustrated above differs from that proposed by
Wierzbicka in that the marking of arguments, not only their expression, is also taken
into account. This definition is also better than that of Wierzbicka, since other
aspects than pragmatic ones are not totally disregarded.

2.4. ‘Multiple definitions’

Up to now, the presented definitions have mainly stressed one aspect of transitivity,
or one facet has clearly dominated. As a result, these definitions have not been able
to analyse the notion of transitivity from different perspectives. For example, the
first presented semantic definitions are useful when studying the typology of
transitivity, but the inadequacies of these definitions may be revealed, if we are to
study transitivity in an individual language (see above). There are also many
definitions of transitivity that try to combine different transitivity features in order
to describe transitivity as exhaustively as possible. The most famous of these is
perhaps the list of ten transitivity parameters proposed by Hopper and Thompson
in their 1980 paper. According to Hopper and Thompson, transitivity consists of ten
different parameters that are presented below (Hopper and Thompson 1980: 252):

HIGH LOW
A. P ARTICIPAN TS 2 or more participants, 1 participant
A and O
B. K IN ESIS action non-action
C. A SPECT telic atelic
D. P U N CTU ALITY punctual non-punctual
E. V O LITIO N ALITY volitional non-volitional
F. A FFIRM ATIO N affirmative negative
G. M O D E realis irrealis
H. A GENCY A high in potency A low in potency
I. A FFFECTED N ESS O F O O totally affected O not affected
J. Individuation of O O highly individuated O non-individuated

As can be seen, the list of Hopper and Thompson consists of parameters some of
which can be viewed as semantic (A, B, C, D, E, H, I), whereas F, G and J are more
38

pragmatic in nature. Syntactic features are not taken into account, since the goal is
to show how the parameters listed above are typically encoded by languages
(Hopper and Thompson 1980:251). Consequently, they cannot have chosen the
structure as their starting point. On the basis of the enumerated parameters Hopper
and Thompson present the following TRANSITIVITY HYPOTHESIS (1980:255):

If two clauses (a) and (b) in a language differ in that (a) is higher in Transitivity
according to any of the features A-J, then if a concomitant grammatical or semantic
difference appears elsewhere in the clause, that difference will also show (a) to be
higher in Transitivity (italics added)

The approach adopted by Hopper and Thompson clearly contributes significantly

to the study of transitivity, since many previously ignored aspects are taken account
of. Hopper and Thompson were also among the first to explicitly view transitivity
as a continuum. This is a clear improvement compared to the purely binary
approaches briefly illustrated above. On the basis of the parameters presented above,
clauses can be distinguished from each other on the basis of their transitivity: the
higher the number of ‘high parameters’ the higher the transitivity of a clause. This
results also in the fact that a structurally intransitive clause can be more transitive
than a two-argument clause (Hopper and Thompson 1980: 254).
Despite the clear advantages and new insights into the notion of transitivity,
the transitivity hypothesis has, however, been criticized. One facet that has been a
target of criticism is the equal importance of all chosen parameters (... according to
any feature...), i.e. Hopper and Thompson consider all transitivity parameters
enumerated above equally important. This is justified in the sense that Hopper and
Thompson explicitly show that all parameters truly contribute to the marking of
transitivity in at least one language. However, there are clear differences in the
typological significance of the parameters. For example, Tsunoda (1985: 393) shows
that parameter I is more important for marking of high transitivity than parameter
H. Hopper and Thompson have also been criticized for their selection of the
parameters. Many selected parameters often co-occur and could consequently be
seen as a single parameter. This is, for example, the case for parameters E and H
(see also Tsunoda 1985: 386). Volitionality is a crucial part of agency, because of
 39

which these parameters most naturally co-vary. The fact that a clause, the agency of
which is higher than that of another clause, is also more transitive as regards the
volitionality of that clause, is tautological and adds nothing to our understanding of
transitivity.
A somewhat similar, yet different definition of transitivity has been proposed
by Lakoff (1977:244). Lakoff proposes the following definition of the prototypical
agent-patient clause:

1. there is agent, who does something

2. there is a patient, who undergoes a change to a new state (the new state is typically
nonnormal or unexpected)
3. the change in the patient results from the action by the agent
4. the agent’s action is volitional
5. the agent is in control of what he does
6. the agent is primarily responsible for what happens (his action and the resulting
state)
7. the agent is the energy source in the action; the patient is the energy goal (that is, the
agent is directing his energies toward the patient)
8. there is a single event (there is spatio-temporal overlap between the agent’s action
and the patient’s change)
9. there is a single, definite agent
10. there is a single, definite patient
11. the agent uses his hands, body, or some instrument
12. the change is the patient is perceptible
13. the agent perceives the change
14. the agent is looking at the patient

According to Lakoff, a prototypical transitive clause has all the features listed above.
The presentation of Lakoff differs from that of Hopper and Thompson in that only
prototypical transitive clauses have been described (furthermore the description is
primarily based on English). Clauses that deviate from those corresponding to the
definition cited above must by default be viewed as less prototypical. As such, it
also makes it possible to view transitivity as a continuum. In distinction from
Hopper and Thompson, Lakoff restricts the notion of (prototypical) transitivity to
events that involve two distinct participants (that can be regarded as agent and
40

patient). Intransitive events cannot be seen as more transitive than transitive ones.
The style of presentation adopted by Lakoff is somewhat more ‘concrete’, since all
properties are explicative. Lakoff’s approach is also somewhat more semantic or
ontological (=non-linguistic), even if the selected properties are more or less similar
to those presented by Hopper and Thompson. However, some of the components
presented above do not have any direct linguistic relevance. For example, features
13 and 14 are very unlikely to contribute to the linguistic marking of transitivity in
any language (they are relevant only as subcomponents of other, more general
properties). As in the case of Hopper and Thompson, also in Lakoff’s definition
some features could be regarded as a single feature. For example, parameters 4-6
and 14 are subcomponents of agency.
The ‘multiple definitions’ examined thus far have been primarily semantic
definitions that have also taken some pragmatic aspects into account. Moreover,
there also definitions that combine semantics with syntax. An example of such a
definition is the following (Testelec 1998:29):

The most elaborate definition I am aware of has been suggested by Kozisnky 1980. To
put Kozinsky's view shortly, a small semantic class of verbs, viz. verbs of destruction
and creation, is assumed to be transitive in its basic voice in all languages. Further, any
verb which requires the same construction(s) as the verbs in the core class do, may be
called transitive.

According to Kozinsky (who is cited by Testelec), the high transitivity of certain

kinds of clause is clearly semantically motivated. Clauses that describe semantically
highly transitive events are classified as structurally transitive (see also Desclés
1998:162). This means that the typical two-argument clause in every language is the
one that is used in the description of highly transitive events (see also Lazard
1998:40). Depending on the language, many other (semantically less transitive)
events can also be described employing the transitive pattern of a given language.
These kinds of definitions can better argue for the high transitivity of (semantically)
transitive clauses than pure syntactic definitions. They are also applicable to studies
of typology of transitivity, since we can base our definition of typical transitive
events on non-structural properties.
 41

2.5. Final remarks

In this chapter, a very brief overview of different transitivity definitions was

illustrated. As can be inferred from the variety of the definitions, transitivity is a
very multilayered notion. The relevant aspects of transitivity can be divided into
three major categories that are semantic, syntactic and pragmatic facets. Of these,
the first and the third are conceptual, whereas the second refers to the linguistic
marking of transitivity (to the ‘output’). All of them naturally make a significant
contribution to the notion of transitivity.
The definitions illustrated in 2.1.-2.4. emphasize different aspects of
(linguistic) transitivity. Those presented in 2.1.-2.3. are based primarily on one of
the relevant parameters, whereas the definitions discussed in 2.4. are combinations
of more than one feature. There are fundamental differences in the nature of these
definitions that are due to the chosen viewpoint. Typical semantic definitions focus
primarily on explaining possible differences without taking into account the
structures to be explained. Typically, semantic definitions only take into account the
transitivity parameters as they are in the non-linguistic world. Consequently, as
already stated in section 2.1. purely semantic definitions do not make it possible to
separate different constructions employed in the description of the same event from
each other. This follows from the fact that different non-linguistic entities can be
linguistically referred to in many different ways without any major effects on the
basic meaning of clauses. As for transitivity, this means, for example, that passive
and active must be both classified as (semantically) transitive constructions. This is
very unfortunate, if we are to describe transitivity in an individual language in which
the distinction between passive and active is the most significant indicator of
(linguistic) transitivity. As a ‘compensation’, however, these definitions are
typologically applicable, since the study of transitivity is not determined by
structural properties of an individual language (the ‘freedom of expression’ is taken
into account). Furthermore, semantic definitions can explain the non-transitive
marking of, for example, experiencer constructions in many languages. They
predict that these events may be expressed differently from semantically highly
transitive events, since they do not correspond to the definition of the typical
42

transitive event comprising an agent and a patient. In order that (purely) semantic
definitions are applicable in this form, they simply must disregard structural
variation.
Pragmatic facets of transitivity differ from semantic ones in that they do not
have clearly definable non-linguistic (or ontological) counterparts. As a
consequence, pragmatic definitions of transitivity can never alone sufficiently
describe transitivity, but they must always be combined with semantics and/or
syntax in order to be at least somehow applicable. For example, effects of pragmatic
aspects on transitivity can be observed only through concrete clauses, since they do
not exist in the non-linguistic world. Pragmatic definitions are also primarily
explanatory in nature. Because of pragmatic definitions we know, for example, that
in some languages, the accusative marking is restricted to definite Patients. The
exact effects are highly language-specific.
Structural definitions are the opposite of semantic and pragmatic definitions,
since they only regard the ‘output’ and disregard the underlying reasons (at least in
the purest form). They are the only kinds of definition that take the clause structure
explicitly into account. Typical syntactic transitivity definitions explain the
transitivity of clauses simply by referring to their structural properties, i.e. the
structure is self-explanatory. Most typically, clauses with a subject and a direct
object are considered transitive, while others are deemed intransitive. Typologically,
these kinds of clause are usually employed in the description of semantically highly
transitive events (see above the citation from Testelec 1998). However, since this
kind of construction can in many languages be employed in the description of many
other kinds of event as well (the transitivity of which can be very low), the
transitivity is usually purely syntactically motivated. Since the motivation is ignored,
structural definitions are very general in nature and they cannot take very many
aspects into account. Structural definitions are further usually based on structural
properties of a certain language, which adversely affects their use in typological
studies.
Semantics, pragmatics and syntax are distinct, yet interrelated facets of
transitivity. Consequently, disregarding two of these aspects unavoidably results in
a somehow insufficient definition (as noted above). In order that our definition
 43

could be more holistic, we should take more of these aspects into account. This is
the case in multiple definitions presented in 2.4. As noted above, pure semantic and
pragmatic definitions are explanatory in nature, whereas syntactic definitions are
purely descriptive, since they usually ignore the rationale behind transitivity. These
faults are largely avoided in the multiple definitions, since they explicitly take more
aspects into account. Especially effective are the kinds of definitions presented by
Kozinsky (cited by Testelec) and Lazard, since they explicitly explain the transitivity
of certain clause types by referring to their (high) semantic transitivity. Since the
basic transitive clause of any language is defined this way, these definitions are
applicable to the study of both individual languages and typology of transitivity.
These kinds of definition also correspond most directly with the approach adopted
here.
3. WHAT IS TRANSITIVITY (HERE)?

In chapter 2, some, more or less widely accepted transitivity definitions outlined by

established authors were illustrated. In addition, their advantages and disadvantages
were briefly discussed. The presentation was not intended as a thorough analysis of
all the relevant transitivity parameters in light of all possible transitivity definitions,
but it served as a mere introduction to the present study. In this chapter, the
theoretical foundations of transitivity are further discussed and developed. The
concept of transitivity is studied from many different perspectives, in addition to
which many different facets of transitivity are discussed in more detail. The
emphasis lies on the interaction of semantics and structure in the explanation of
transitivity phenomena. Of these two, the semantics is primary, because of which
we begin our presentation with a detailed analysis of semantic transitivity and
different aspects associated with it. The existence of two distinct participants that
clearly bear different semantic roles is here regarded as the most important
transitivity feature. The notion of high semantic transitivity is illustrated in light of
the prototypical action. This criterion is not sufficient, but we have to take other
features into account as well. These include ‘ontological’ vs. ‘conceptual’,
‘semantic’ vs. ‘valencial’ and ‘structural’ vs. ‘interpreted’ transitivity.

3.1. Some relevant notions

Before proceeding to a detailed analysis of a prototypical transitive action we

illustrate some notions that are relevant for the discussion below.

3.1.1. Ontological vs. ‘conceptual’ transitivity features

The first distinction made here is that between ontological and conceptual
transitivity features. In general, features of ontological transitivity are here
understood referring to the transitivity of events in the non-linguistic world. The
label ‘ontological transitivity’ refers to events in the non-linguistic world ‘as they
are’. The relevant aspects have an independent existence of their own in the non-
linguistic world (even if they are not being observed by animate entities) and they
are salient to all of us. So for ontological features of transitivity, it is not relevant
 45

how the event is conceptualized, but despite different conceptualizations the

ontological transitivity of the event remains the same. The most relevant ontological
feature of transitivity is the number of participants involved in an event and the
semantic roles associated with them. Ontologically, (two-participant) events that
result in a clear change of state of the patient are transitive, while those that do not,
are less transitive or even fully intransitive. Ontological transitivity can either refer
to events as they are, including all possible properties (irrespective of how irrelevant
they are for the differentiation between events and for transitivity) of the event in
question or the concept can be understood in a more restricted sense. In the first
sense, all events differ somehow from each other as regards their transitivity (the
differences can be very insignificant in this case), since all events in the non-
linguistic world only have a single occurrence that differs somehow from all others.
For example, the nature of the patient affectedness varies; the patient is clearly
affected differently as a result of ‘painting’ than as a result of ‘washing’. However,
these kinds of difference are usually irrelevant as regards the transitivity of events.
Hence, the term is best employed in a more ‘restricted’ sense. This means that all
insignificant properties are ignored and the ‘true’ transitivity of events is focussed
on (this includes only the most relevant facets of transitivity, see below). In this
view, ontological transitivity can be understood as some kind of basic, general
meaning of events. The basic meaning refers, for example, to our knowledge about
the semantic roles of participants partaking in different events and also about the
nature of the event itself (is it, for example, affective or effective). On the basis of
the semantic roles of participants, we are able to distinguish between events
according to their transitivity. For example, ‘murdering’ is clearly a transitive event,
since it involves a highly agentive agent and a directly affected patient. On the other
hand, ‘seeing’ is clearly a less transitive event, since there is no agent nor patient
present. Ontological transitivity is not a clear-cut concept, but the boundaries
between different events are more or less fuzzy. Hence, ontological transitivity is
best viewed as a continuum.
Ontological transitivity (as for linguistic manifestation of transitivity) is best
defined as our idea about different events in the non-linguistic world. Based on the
recurrence of events, we are able to make generalizations about their relevant
46

properties. Only the bare nature of events is relevant is this respect. This information
is employed in the description of events and in the interpretation of constructions.
The features of ontological transitivity are usually absolute in nature and the
ontological information about the nature of events is common for all language users
(regardless of the language they speak). The absolute nature of these features means
that we all are able to distinguish ‘killing’ from ‘hearing’ and we all agree on this
distinction (provided that we behave rationally). The latter refers to the common (or
social) nature of these features. The most relevant features of ontological transitivity
are the semantic roles associated with different events (along with the nature of the
event itself).13 We know that different events involve different kinds of entity that
bear certain semantic roles (that might be decomposable into different sub-features
as suggested, for example, by Dowty in his 1991 paper, these sub-features are
usually also salient). For example, ‘hitting’ involves someone who is doing
something (an agent) and someone or something who/that is the target (a patient)
of the action. The information about the nature of events is faith in the sense that the
ontological transitivity remains unaffected, even if the linguistic manifestation
changes (this naturally requires that we are not dealing with a transitivity alternation
that is conditioned by changes in ontological transitivity). For example, we know
that ‘eating’ requires two participants, even if they are not expressed explicitly. We
cannot make the event ontologically intransitive by omitting the patient. Since the
ontological transitivity features are inferred from non-linguistic events, they are
relevant in the separation of distinct events (or event types). Changes in ontological
transitivity always result in changes in the basic meaning of the given event.
The (linguistic) transitivity of events is based primarily on features of
ontological transitivity (and to the linguistic description of any event in general),
since this level refers to the events we want to describe. The ontological level cannot
be ignored. However, many non-ontological facets also contribute significantly to
the transitivity of clauses. These kinds of feature are here referred to as conceptual,
since they present different conceptualizations of the same events. Basically,
conceptual transitivity is based on our (differential) observation of the same events.

13
Semantic roles are here understood as our ideas about the nature of participants in events.
 47

Since we are all individuals, we observe events somewhat differently (which does
not affect the nature of events in any way), emphasize different aspects of events,
and are also able to describe events from different perspectives. Relevant properties
are focussed on and accorded more linguistic prominence. Despite the differential
conceptualizations of events, their ontological transitivity (in the sense adopted
here) remains unaffected. As opposed to the absolute and invariant ontological
features, conceptual features are always relative, individual and variable. This means
that there are no absolute or constant criteria of conceptual transitivity, but we all
conceptualize events differently and the conceptualization varies in different
situations. For example, someone may think that the agent in an event is the more
relevant participant, whereas someone else chooses to focus on the patient. There
is not a single ‘correct’ way of conceptualizing events, while ontologically this
usually is the case (if we behave rationally). This is due to the fact that features of
conceptual transitivity do not exist non-linguistically, as do typical ontological
features. They can typically be observed only through differential marking of
clauses. While features of ontological transitivity separate different events from each
other, different conceptualizations usually impinge on a single event. For example,
the ontological transitivity of an event remains unaffected, if we choose to focus on
the patient instead of agent.
Conceptual transitivity comprises (in the sense adopted here) all the aspects
not readily and directly inferable from the ontology of events. Relevant aspects
include differential pragmatic weighting of participants (degrees of definiteness),
degree of agency, aspect and Aktionsart. The first of these is clearly the most
conceptual one, whereas others include some ontology as well (as will be shown).
Consequently, these features differ from each other in what kinds of effect they have
on the transitivity of clauses or events and how they are motivated. For example,
different pragmatic weighting of participants and its consequences for the structure
of clauses has no effect whatsoever on the ontological transitivity (the event itself
remains exactly the same): e.g. passive and active are both describing the same
event, but from different perspectives. Changes in the agency or aspect, for their
part, also affect the basic meaning of events, even if the basic event itself remains
more or less the same. For example, the event ‘he broke the mirror accidentally’ is
48

not prototypically transitive, since the action of agent is not volitional and the full
control is lacking. However, since we are also in this case dealing with somewhat
different conceptualizations of ‘breaking’, we may classify these relevant transitivity
features as conceptual. There are no absolute, ontological criteria of agency, but the
degree of agency varies and we all interpret this concept somewhat differently
(perhaps because we emphasize different aspects of it). In both highly agentive and
less agentive ‘breaking’, someone is doing something that results in a salient
change-of-state in the patient, i.e. the basic event is the same. That there is also
some semantics involved may be illustrated by the fact that clauses he broke the
mirror and he broke the window accidentally cannot be arbitrarily chosen to
describe a breaking scene, but there are semantic (or ontological) differences
involved. It depends on the conceptualized degree of agency, which structure we
choose to employ. The first clause is appropriate, if we think (or know) that the
agent is intending something to happen and is hence a typical agent. The other
construction is used, if the degree of agency is (or we wish it to be) lower. The
clearly conceptual (i.e. non-absolute) nature of agency is also reflected in that in
many languages markers that relate to the expression of agency are possible only if
the Agent is in the first person (examples include Lhasa Tibetan (see DeLancey
1984:132), Tucano and Tariana14). Aspect and Aktionsart are rather similar to
features of agency as regards their nature as parts of conceptual transitivity. Neither
has a clear non-linguistic equivalence, but it always depends on our judgement,
which facet we focus on.
Even if ontological transitivity can be understood as a universal notion in the
sense that every speaker of any language can distinguish between ontologically
different events, languages differ crucially from each other in how explicitly they
differentiate between events based on their ontological transitivity. Before
proceeding to the next topic, it should, however, be pointed out that these kinds of
difference are not due to differences in perception of the same events, but they
follow from differences between distinct events. Even if language A employs a
typical transitive construction in the expression of experiencer constructions (e.g.

14
The last two languages are taken from Timothy Curnow’s presentation on typology of evidentials
at ALT IV in Santa Barbara July 20, 2001.
 49

‘he sees me’), whereas in language B the experiencer appears, for example, in the
dative, the event as such is the same. The dative marking of experiencer in language
B is due to the less transitive nature of experiencer constructions in comparison with
ontologically highly transitive events. The structure of a language simply allows a
more explicit marking in language B. The differences are ontological, not (merely)
conceptual (there might be some differences in the conceptualization because of
which we wish to avoid saying that the perception is fully identical). We are dealing
with genuine conceptual differences, if the structures are distinct, but the profiled
events the same.

3.1.2. Transitive valence vs. semantic transitivity of events

As noted above, typical transitive events involve two participants, one of which
instigates the event, whereas the other is targeted by it. Put roughly, the first of these
criteria is here defined as transitive valence and the latter gives a very simple picture
of semantic transitivity as it is understood here. Transitive valence refers here to the
number of participants partaking in events. Transitive valence is in principle a
notion with clear boundaries, since the valence is always one (as in ‘he runs’), two
(‘he hits me’) or three (‘he gives me a book’). There are no intermediate forms.
However, the distinction between a participant and a non-participant (a circumstant)
is not that easy to draw, which makes the overall picture less clear. For the sake of
convenience, we have adopted a rather simplistic approach to this problematic
notion. As a participant is regarded any entity that inheres in the semantics of a
given event and without which the event is not ‘complete’. Everything else is
regarded as a circumstant (i.e. a non-core participant). Typical participants are
therefore agents, patients and experiencers, while locative adverbials are typical
circumstants. This simplistic definition does not cover every possible case, but
should suffice for the purposes of the present study. An event is here regarded as
transitive only if ttwo (or more) participants are involved and semantic transitivity
is deemed relevant only if the valence is two (or higher). This follows from the
nature of semantic transitivity: semantically (or ontologically), transitivity is
typically defined as an efficient energy flow from one participant (agent) to another
50

(patient). In order that an efficient energy flow of this kind is possible, there must
naturally be at least two participants present. The presence of two participants is,
however, not alone a sufficient criterion for high semantic transitivity, since the
relation between participants may be very intransitive in that no carrying over of an
action takes place. In order that an event be justifiably called transitive, certain
semantic parameters must be present as well. These are here classified as features
of semantic transitivity. Typical semantic parameters include features of agency and
affectedness of the patient. In the following, these two aspects of transitivity are
briefly illustrated in order to show that both of them make a significant contribution
to linguistic transitivity.
As regards the transitive valence of events, an event is here classified as
transitive in all the cases in which there are two participants involved irrespective
of their semantic roles (it has to stressed, however, that this is not alone a sufficient
criterion for high transitivity). Furthermore, the two participants must stand in an
asymmetric relation, which means that one of them is doing something and the other
is somehow affected by this. Events in which two participants are involved in the
same intransitive action are naturally excluded. For example, an event involving two
agents walking together is not transitive. The involvement of participants must also
be required directly by the semantics of the event denoted.15 The transitive valence
refers to the number of core participants. This means that only roles of participants
inherent in the semantics of the event are relevant. The participants that are included
in the transitive valence are always integral parts of events, and they cannot be
eliminated without this resulting in obvious semantic changes in the basic structure
of events. Peripheral participants are excluded from the notion of transitive valence.
For example, instruments (e.g. ‘he killed me with a knife’) are here not regarded
as parts of transitive valence, irrespective of how significant their contribution to the
nature of an event might be (for example, ‘rope’ is not viewed as a part of the
transitive valence of ‘tying’, even if we cannot tie anyone without some kind of
rope-like object). All participants that are included in the transitive valence of events
must further exist independently in the non-linguistic world. The participants may

15
This excludes applicatives based on intransitive clauses.
 51

exist prior to the event (as in ‘he hit me’ or ‘he gave me the book’) or the existence
may be a direct result of the event (e.g. ‘he wrote a book’). Furthermore, we must
be able to distinguish between the participants. The participants cannot be merely
(conceptualized) products of certain events, whose existence directly depends on
these events. For example, cognate objects (e.g. he is singing a song, they are
dancing tango) are here not viewed as concrete participants in events, since their
existence depends on the event they are parts of. We cannot here distinguish
between the event and its target. The product exists only as long as the event
proceeds. Hence, cognate objects are not real products of events. Non-linguistically,
these events are intransitive in their valence.
The features of semantic transitivity are related to the nature of events. Chief
among these are different features of agency of the instigator and affectedness of the
patient. Other features include, for example, aspect, kinesis and punctuality. The
features may be of two kinds that roughly coincide with the features of ontological
and conceptual transitivity discussed in the previous section. These are in this
section referred to as inherent and variant features. Inherent features refer to the
semantic roles of the participants in events. Each event is characterized by a certain
inventory of semantic roles. For example, semantically highly transitive events
involve an agent and an affected patient, whereas experience constructions involve
an experiencer and a stimulus. These roles are integral parts of the nature of the
events both of which are transitive in their valence, but only the first kind of event
can be considered semantically transitive. Only events whose participants can be
viewed as agents and patients are semantically (highly) transitive. Semantic
transitivity of inherently transitive events is, however, not absolute or constant, but
semantic transitivity can vary. This is where the variant transitivity features make
their contribution. For example, the transitivity of the inherently highly transitive
event ‘x kills y’ can vary according to the degree of agency of the instigator. In the
most transitive cases, the agent causes the change-of-state in the patient volitionally.
It is, however, possible that the instigator of the event is only indirectly responsible
for the event. This can be due either to the (inherent) nature of the entity in question
(e.g. ‘six people were killed in the storm’) or to situational changes in the agency
(‘they caused the man to die (e.g. by not giving him anything to eat)’). The
52

transitivity of these events is not so high as that of cases in which the action is
purposefully directed at the target.
As has been noted, transitive valence and semantic transitivity are independent
of each other. An event (and consequently a clause) can be transitive in valence
without being semantically transitive. The semantic transitivity features of an event
are not inferable from the transitive valence of events, but they are completely
independent of it. Experiencer constructions are very good examples of events this.
They inherently involve two participants (experiencer and stimulus), even if there
is no semantic transitivity involved (in the sense adopted here). The independence
of transitive valence and semantic transitivity is also reflected in the fact that an
event intransitive in valence can outrank an event transitive in valence in transitivity
(cf. Hopper and Thompson above). This is due to the nature of the relevant semantic
transitivity features. Features of agency, affectedness of the patient and other
features of semantic transitivity are not properties of transitive events only, but all
of these may be present in intransitive events as well. For example, ‘he is running’
is a highly agentive action. In this respect, it is semantically more transitive (i.e.
involves more features of high semantic transitivity) than, for example, experiencer
constructions that are transitive only in their valence. However, there is no
transmission of energy between participants involved in either of these events.
Transitive events differ from intransitive ones, since only in transitive events both
agentive and patientive features are present at the same time. As a result, only events
that are transitive both in their valence and on the basis of their semantics are
regarded as transitive events in this study.
The independence of transitive valence and semantic transitivity is further
reflected in the fact that an increase in transitive valence does not necessarily result
in an increase in the semantic transitivity of an event. In similar vein, semantic
transitivity may be totally unaffected by a decrease in transitive valence. A typical
example of the former case is the causativization of transitive events. In this case,
the original event involves two participants and the resulting one three, but there is
no increase in any semantic transitivity feature of the causativized event. If anything,
the effect is the opposite (see ch. 5.2.5.).
Both transitive valence and semantic transitivity are typologically relevant
 53

features as regards the transitive marking of clauses. In most languages, the

transitive valence is the more important of these two in this respect. The transitive
valence is the basis for transitive marking in general. Only if an event involves two
independent participants, can the clause pattern employed in the description of the
event be transitive. In languages in which the transitive valence is clearly the
primary of these two, most of the events that are transitive in their valence are
described using some basic transitive structure. The effects of semantic transitivity
on the marking of clauses are more indirect in nature. In languages in which the
transitive valence (of two) is not a sufficient criterion for high transitivity (which is
the case in most languages of the world), we can justify the morphosyntactic
transitivity of clauses only by referring to semantics. This means that only clauses
that are semantically highly transitive can be regarded as structurally transitive (cf.
Kozinsky and Lazard above).

3.1.3. Interpreted vs. structural transitivity

The relation of transitive valence and structural transitivity is in principle iconic.

This means that the number of explicitly marked arguments coincides with that of
participants, i.e. events involving two participants can be expressed using two-actant
constructions.. However, this statement refers only to prototypical cases. The
number of explicitly expressed arguments is conditioned by other factors as well.
This often results in a disagreement between semantic and structural transitivity, i.e.
the number of arguments differs from that of participants. In other words, the
structural and ‘interpreted’ transitivity disagree. An event that is transitive in
valence may be described by an intransitive construction. Structurally, the given
clause is intransitive, but we interpret the event as transitive based on the semantics
of the verb. It goes without saying that the iconicity of structural transitivity and
transitive valence does not hold in these cases. For example, we do not interpret the
clause he is eating as referring to an event that is intransitive in valence, even if the
structure is identical to that employed in the description of intransitive events (e.g.
he is running). Despite the omission of the Patient, we know that the event profiled
54

is transitive, since there is no eating without something eaten. Hence, transitivity is

determined by semantics, not by the number of explicitly expressed arguments.
In cases like the one illustrated above, the omission of arguments is only
structural. The basic transitivity of the event remains unaffected. Typically, the
number of participants involved is higher than that of arguments. Arguments
considered irrelevant for the intended interpretation are often omitted. It is not
necessary to refer to a participant explicitly, if its nature can be inferred otherwise.
For example, the omitted Patient in he eats must refer to something that is usually
eaten (fish, meat, vegetables etc.) not to an unusual patient such as paper or plastic.
Typical patients are in a way a conventionalized part of the meaning of verbs, which
makes the omission of Patients possible (there is no need to explicitly express
something that is already known to us).
So far, we have exemplified cases in which the disagreement of syntactic and
interpreted transitivity follows from the inherent transitive valence of events. This
kind of disagreement is possible, since we can infer the number of participants from
the semantics of the verb. The number of participants is, naturally, not the only
information that inheres in the verb, but verbs contain information about the nature
of the event as well. The semantic roles of participants make a very significant
contribution in this respect. The inventory of semantic roles is an integral part of the
meaning of verbs. Identically to the number of participants involved, we infer the
semantic roles associated with the arguments from the semantics of the verb
regardless of their morphosyntactic form. For example, the Agent in he kills me is
interpreted as an agent, whereas that of he sees me is construed as an experiencer,
even if the morphosyntactic marking of the arguments is identical. In similar vein,
me is interpreted differently in these cases irrespective of the identical
morphological marking (accusative marking does not necessarily mean that the
participant would somehow be affected by the action). Also in these cases, structural
and interpreted transitivity disagree. Both exemplified constructions are examples
of basic transitive clauses of English. Consequently, the construction that can justly
be viewed as the basic transitive construction is also employed in the description of
 55

less transitive events.16 If our interpretation of events was directly based on

morphosyntax, we would interpret both of these events as transitive, even if the
differences in semantic transitivity are very obvious. Also in this case, the semantics
makes a more significant contribution to the interpretation of clauses. The structure
of clauses, for its part, is determined by the transitive valence alone (if the
arguments are explicitly expressed), which causes the kinds of disagreement
illustrated. Because the semantic roles of participants are an integral part of the
semantics of the verb, it suffices to refer to participants without specifying their
semantic roles in any way structurally. An explicit marking of arguments on the
basis of their semantic roles is not important as regards the reading of clauses.
So far, we have been dealing with cases in which the structural transitivity of
clauses has been rather irrelevant, since we have been able to infer the number of
participants along with their semantic roles directly from verb semantics. The
omission of arguments and the identical marking of semantically very different
arguments is possible, since the expression or explicit marking of arguments would
only emphasize aspects of events that inhere in the semantics of the verb. The
argument omission and their identical marking is in both cases conditioned by the
economy principle of language use. Information that can be inferred non-
linguistically or differences that are easy to recover are left unmarked (see, e.g.
Kibrik 1985: 271). Conversely, an explicit marking is necessary, if the intended
meaning is not assured without it. This is the case, if the transitivity of an event
differs from that inherent in the basic semantics of the verb. As regards the
structural and interpreted transitivity, the relation must be iconic here, since we
construe the event based primarily on linguistic cues. For example, in English the
patient is usually affected as a result of shooting. If this is the case, the object
argument appears in the default form, as in he shot me. However, if we want to
emphasize the lower transitivity of the given event, we must resort to linguistic cues.
The clause he shot at me is explicitly marked for its lower transitivity and we
consequently interpret the event as less transitive. This kind of marking is necessary
for the right reading, since the meaning differs from the expected one. The marking

16
This occurs in many other languages as well, it is not typical of English only.
56

is in these cases by default iconic, since the less transitive marking is based
primarily on the ‘non-typicality’.

3.2. On semantic transitivity

3.2.1. On relevant criteria

A typical transitive event can generally be defined as an event that involves two
participants (agent and patient) that stand in an asymmetric relation, i.e. one of the
participants instigates the event, while the other is directly and in a salient manner
affected by it. The state of the patient before and after the event must be
distinguishable. Agency of the instigator and direct affectedness of the patient (along
with an efficient transfer of energy from agent to patient) are here viewed as the
most relevant features of semantic transitivity. If either or both of these is/are
lacking, we are not entitled to speak of a transitive event. This kind of event is
viewed as typically transitive already in the early stages of first language acquisition
(see Slobin 1982). The relevant concepts of transitivity are neither clear-cut nor
single layered features, but they are more continuum-like and consist of many
(usually somehow interrelated) subparts. Moreover, there are also other properties
that do not relate directly to agency or affectedness parameters, but more generally
to the overall nature of prototypical transitive events. These will all be dwelled on
in what follows.
As noted, transitive events inherently involve two concrete, distinguishable
participants. This means that the semantic role assignment is obvious. In typical
cases, there are no overlaps, but the agent is alone responsible for the agentive
features and the patient for features associated with the affectedness parameter. In
other words, the agent is the active participant in the event, while the patient is
merely a passive target. Participants in a transitive event ‘share the responsibility’
for its nature. In this respect, transitive events differ from intransitive events in
which all the relevant facets are completely related to a single participant (that might
be either more agent-like or more patient-like in nature).
In typical transitive events, the action by the agent is directly targeted at the
 57

entity that gets affected as a result of the action. Furthermore, the agent is intending
something to happen and is aware of the consequences of his/her action. This has
the consequence that typical transitive events must be construed as single events in
which the action by one participant immediately, directly and in a perceptible
manner results in a change-of-state of the other (cf. Lakoff above and also Desclés
& Guentchéva 1993:13). There must not be any longer interval between the action
by the agent and the resultant state of the patient. In the latter kinds of case, it is not
clear whether the resultant state actually followed from the action of the agent.
Furthermore, we cannot be sure whether the patient is a part of the event in which
the original agent partook. A typical transitive event must not involve two separate
events that are somehow interrelated and could consequently be conceptualized as
a single event. Examples of this are provided by causatives that involve three
participants (causer, causee and patient). These can be regarded as consisting of two
separate events involving two participants. The first event involves a causer and a
causee and the latter a causee and a patient. The latter event follows from the first
one. The interval between the events may be very long: the first event does not
necessarily result in the second immediately. Consequently, we are not necessarily
dealing with a single event in cases like this.
Typical transitive events are also concrete and salient in nature. We must have
direct evidence for their occurrence . Hence, for example, events that are not taking
place as we speak are not examples of typical transitive events. The salience feature
means that the action of the agent must be readily observable. Put concretely, this
refers to events in which the agent is using some part of his body or some instrument
to impinge on the state of the patient (cf. Lakoff 1977: 244). Only events that
involve concrete actions can be considered typically transitive. Events involving, for
example, perception are not examples of typical transitive events, since we cannot
readily state what is occurring, in addition to which the semantic role assignment is
less than clear.
Transitive events (defined as an interaction of an agent and a patient that
eventually results in a salient change in the state of the patient participant) may be
said to consist of four different phases that are all relevant for high transitivity, even
if these differ crucially from each other as regards our observation of transitivity (see
58

below). A typical transitive event comprises all of the relevant phases. The first
phase is the planning of the event. As noted above, the agent of a typical transitive
event is acting volitionally and is in control of the event, which implies that the
action must be planned somehow. At this point, the only relevant participant is the
agent that must exist independently of and prior to the event. If the agent chooses
not to proceed, no event will take place. The second phase is the initiation. At this
stage, the agent initiates the event, i.e. it is doing something that will eventually
result in a transitive event. In typically transitive cases, the initiating act is
intentional and it coincides with an act that typically results in events that are being
initiated. This means that the agent is aware of what s/he is doing. Also at this stage,
the agent is the chief participant. If the agent ceases to act, the event will usually not
take place. The patient is, however, also involved at this point, since a successful
initiation requires that there be some kind of patient present that will be targeted at.
The third phase refers to the event itself. The third stage is the most directly and
saliently observable one, since both participants are relevant and ‘activated’. The
third stage refers to what is usually understood as a transitive event. However, even
if an agent is acting on a patient and we are able to observe that directly, we cannot
be sure, whether the action will be completed successfully or not. Because of this,
the fourth phase also has to be taken into account when defining high transitivity.
This phase refers to the result of the event. The event is no longer occurring and we
are able to observe it only through the resultant state of the patient. The action of the
agent can no longer be observed and it is relevant only, since it has led to a change
in the state of the patient.
In the ‘traditional view’, we speak of high transitivity at the third stage, since
only then are both participants involved in the event. At the two first stages, the
agent is the only relevant participant and there is no transitivity involved. The
eventual event is intransitive both in its valence and based on its semantics. If the
intended event remains a mere intention in the agent’s mind, we cannot consider the
event transitive, since nothing has occurred. As regards the resultant state, the event
is not transitive any more, since the agent has ceased to act and we are dealing with
a mere state of an entity. The number of participants suffices to make the event
intransitive. However, even if transitivity in a strict sense is restricted to the third
 59

stage, all four phases make a significant contribution to the transitivity of events.
Deviations from the scheme illustrated above at any of the stages can result in a less
than perfect transitivity of events (and consequently clauses). For example, events
not planned are in many languages marked differently from planned ones. These
aspects are all discussed in more detail in section 5.3.1.1.1., because of which one
example must suffice for now.
Agent and patient make a very different kind of contribution to transitivity of
events. Agent is the participant that is primarily responsible for the occurrence of
the event (cf. above). Agent is also the only actively acting participant in a typical
transitive event. Consequently, agent may be construed as the active part of
transitive events. If the agent chooses not to act, no event will take place. On the
other hand, agency alone is far from being a sufficient criterion for transitivity (see
above), but the action has to be targeted at an external participant. We construe the
agent as the active part of transitive events, while the patient is a passive participant
not responsible for the occurrence of events. Furthermore, the participation of
patients in events is usually not volitional. Patients do not usually choose to be
targets of actions, whereas agents usually partake in events volitionally. As for
linguistic transitivity, this means that only non-volitional participation of agents can
result in a transitivity alternation, while volitionality related to patients is irrelevant
in this respect. The patient simply registers the result of the action without being
able to control it. The activeness vs. passiveness of the relevant participants is also
reflected in the fact that humans are the most typical agents, whereas patients are
typically inanimate entities. Inanimate entities cannot be construed as typical agents,
whereas humans can also be targets of actions. Inanimate entities are not volitional,
while being a passive target is independent on animacy.
As integral parts of high transitivity, agency and affectedness are completely
independent of each other. High agency does not imply affectedness of the patient
or vice versa. This is best illustrated by the fact that both may alone be features of
intransitive events. Also in the case of transitive events, the degree of agency may
change drastically without this resulting in any changes in the affectedness of the
patient or vice versa. For example, the patient is totally affected in both ‘he broke
the window (purposefully)’ and ‘he broke the window accidentally’ despite the
60

changes in agency. This also has clear consequences for transitivity alternations. In
case these two parameters would always co-vary, the number and nature of different
transitivity alternations would be much lower and more ‘prescribed’ than is the
actual case. There would, for example, not be transitivity alternations that affect the
agency parameter only.
Parameters of agency and affectedness are also motivated differently and they
are related to different aspects of events. Agency does not usually depend on the
inherent semantics of events, whereas affectedness of the patient is usually
conditioned by it (or the affectedness of the patient determines the transitivity).
Hence, the agency may vary within one and the same basic event from very high
(e.g. ‘he broke the vase purposefully’) to very low (e.g. ‘the wind broke the vase’).
Furthermore, high degrees of agency are not related to certain events only, but the
agency of different events may be the same. Certain kinds of event tend to be
agentive, but the degree of agency is not conditioned by the semantics of events
only. Agency comprises such features as volitionality and control. The exact nature
of what the agent is doing is irrelevant. As regards the affectedness parameter, on
the other hand, every event implies a specific kind and degree of affectedness from
the target. All events differ from each other somehow in the nature and degree of
affectedness of the patient. The relation between the nature of specific events and
the target is much closer than that between the nature of agent and events (see also
Comrie 1982:112ff and Keenan 1987:171f).
As briefly noted above, agency and affectedness parameters are not binary
features that consist of one feature only. They are both continua based on a number
of features. Relevant subcomponents of agency include volitionality, intentionality
or purposefulness, control and unaffectedness. The most typical agents have all of
the properties enumerated here. The feature of volitionality means that the agent is
not forced to act, but he chooses to do so because of his/her own free will. This also
implies that the agent has the option not to act. Intentionality or purposefulness are
closely related to volitionality. Intentionality refers here to the fact that the action
performed by the agent is the one he/she wanted to perform and that the entity
targeted at is the intended one. The agent must also employ means appropriate to
achieve the wished result. Furthermore, a typical agent also controls the event it is
 61

partaking in. He/she knows what is happening and he/she could stop acting.
Unaffectedness is here viewed as an important part of agency, since the participant
affected as a result of the event is the patient and not the agent. If the agent gets
affected, it becomes more patient-like, which adversely affects its agentivity.
Furthermore, in this case the features of affectedness are distributed, which is not
the case in typical transitive events.
As illustrated above, the agency is a very multi-layered notion. This is
primarily due to the active and human nature of agents. Human behaviour is in
general a very complex phenomenon and agency is no exception in this respect. On
the other hand, since the patient is a passive participant in transitive events,
humanness is irrelevant as regards the affectedness parameter.17 Human and
inanimate participants are both equally capable of being passive. The relevant
features of affectedness include the degree and manner of affectedness. The degree
of affectedness refers here to whether an event causes a salient change of state in the
target or not. For example, the patient of ‘he killed the man’ is unarguably affected,
whereas the target in ‘he saw the house’ is not. There are, however, also many
unclear cases. The manner of affectedness also makes a very significant contribution
to linguistic transitivity because of which it is here regarded as an integral part of
semantic transitivity. One facet of this has already been discussed, i.e. whether the
patient is saliently or non-saliently affected. As noted previously, typical transitive
events are salient in that they can be directly observed. Hence, the affectedness of
the patient participant (that manifests the effect of the event) must also be very
salient. Targets of breaking of all kinds are excellent examples of saliently affected
patients. The affectedness of the patient is far less salient in the case of mental
affectedness. Examples of this include cases like ‘he hurt her mentally’ or ‘she
drove him crazy’. In these cases, the patient does not undergo any physical change
of state. However, we are dealing with an affected patient, since the mental state of
the patient has changed (more or less drastically).

17
Even if humans are capable of perceiving the effects of actions more directly.
62

3.2.2. On the nature of transitive events

In the previous section, we discussed the notion of the basic transitive event. In this
section, we will broaden the scope by taking account of the motivation of transitivity
in different events. Despite the more or less identical transitivity features of the
events at issue (i.e. all the events could be considered transitive in light of the
definition given in the previous section), the structure of these events is clearly
different, which has clear consequences for their linguistic transitivity. For example,
as will be shown in (412)-(430), the nature of differently motivated transitive events
affects the nature of transitivity alternations applicable to given events.
Consequently, a brief presentation of these types is in order here. The significance
of the motivation for transitivity alternations will not be discussed in any detail, but
the present section serves merely as a basis for the upcoming discussion. We begin
our presentation by illustrating differences between affective and effective events
in detail, after which we proceed to differences between inherently transitive,
causative and ‘extended intransitive’ events. There are also certain differences in the
semantic transitivity that will be touched upon, even if all the events discussed will
be considered transitive.
To begin with, it is appropriate to emphasize the independence and uniqueness
of transitive events. This means that they are not merely combinations of different
intransitive events, but they are an event type of their own. Transitive events involve
both an agent and a patient, which distinguishes them from intransitive events that
only comprise either. Since intransitive events involve either of the relevant parts
of transitive events, transitive events could be regarded as combinations of different
agentive and patientive events that together constitute a transitive event. In this
view, intransitive events would be primary and the basis for other kinds of event. As
is evident, this kind of combination of intransitive events does not result in highly
transitive events. Transitive events constitute clearly an event type of their own that
is only indirectly related to intransitive events. First, as noted earlier, typical
transitive events are single events with two participants. They are not combinations
of two events both of which involve one participant. In the case of highly transitive
events, the very nature of the event itself requires that an agent and a patient be
 63

involved. Second, the nature of the action of the agent and the affectedness of the
patient are conditioned by the non-linguistic nature of a given event. This means that
the agent must adjust its action according to the desired result. For example, we
(usually) cannot break a window by running back and forth in front of it.
Occasionally, a combination of two intransitive events can be regarded as a
transitive event, but these are not typical or constitutionalized instances of transitive
events. The semantic transitivity of these events is always somewhat reduced. For
example, a child that is running around in the house can (accidentally) break a vase.
However, running is not a typical way of breaking things nor does the breaking
obligatorily involve running. In transitive events, the nature of participation is more
directly (but not exclusively) determined by the semantics of events.
As stated in the previous section, typical transitive events result in a salient
change-of-state in the patient. The event has no effect on the existence of the
participants per se, but the obvious result is the salient change-of-state. For example,
in the case of breaking the patient-to-be is unbroken before the event and is broken
after the event has occurred. In both cases, the entity in question exists in some
form. Following the established terminology, these kinds of event are here referred
to as affective events (cf. e.g. Lyons 1977:491 and Hopper 1985:70ff). Typical
examples of affective events include events like ‘the man killed the bear’ and ‘the
child broke the window’. Affective events are thus transitive events that involve an
efficient transfer of energy from one participant to another that results in a state-of-
change in the patient participant. Agents of affected events direct their actions at
patients in order to manipulate their state (but not their identity). The other kind of
transitive event is illustrated by events that result in a creation of an entity that did
not exist prior to the event. These events are here referred to as effective (also
following the established terminology). Effective events involve two participants the
action by one results in the creation of the other. Typical effective events are
consequently the likes of ‘he built a house’ or ‘the woman wrote a book’. As
opposed to affected patients, the identity of the entities in question changes
radically. For example, a house is a considerable amount of wood, bricks and
concrete prior to the successful building event. If manipulated appropriately, these
materials can be transformed into a house. The action by the agent is directed at the
64

creation of this entity, since it cannot, naturally, be targeted directly at the patient,
because it does not exist yet. The nature of the energy flow is therefore different (for
a different marking of affective and effective events (in Akan and Ga) see Hopper
1985:75).
In addition to the affective vs. effective distinction, we can distinguish three
kinds of transitive event based on how the transitivity is motivated. The differential
motivation refers here to the way the number of participants in the event is
motivated. On the basis of this, we may divide the events into inherently transitive,
causative and ‘extended intransitive’ events.18 As inherently transitive are here
regarded events that obligatorily involve both an agent and a patient role.19 The label
is not restricted only to highly transitive events, but the label comprises also events
that are inherently transitive in their valence. Naturally though, events like
‘washing’ and ‘painting’ illustrate more typical instances of inherently transitive
events than ‘seeing’ and ‘hearing’. In these cases, it is not possible to omit either
participant, since the event is characterized by them both. There are no underlying
intransitive events from which the inherently transitive events could be derived.
Examples of inherently transitive events are ‘eating’, ‘building’ and ‘washing’.
Causative events are originally patientive intransitive events that are caused by an
external agent. Consequently, causative events are in a way based on underlying
intransitive events.20 Examples include ‘killing’, ‘breaking’ (transitive) and ‘felling’.
Extended intransitives are similar to causatives in that (as the term implies) we are
dealing with a transitive version of possibly intransitive event that has been
transitivized. Extended intransitives differ from causatives in that the added
participant is a patient and the original event involves an agent. Examples include
‘chasing’, ‘following’ and ‘spitting on someone/something’. Extended intransitives
are also somewhat less transitive than the events of the two other kinds. All of these

18
The division has been influenced by Halliday’s division into ‘transitive’ and ‘ergative’ events (see
also Davidse 1992), but is not identical with it, extended intransitives are not used in the structural
Dixonian sense, cf. Dixon & Aikhenvald 2000b:3).

19
The term is used in a radically different way from that of Payne 1985 who defines inherent
transitivity more structurally by referring to the number of transitivity alternations possibly applied.

20
W e do not wish to claim that intransitive events are the primary ones.
 65

types and their properties are discussed in more detail in what follows. The
presentation follows the order the events were introduced.
The occurrence of inherently transitive events relies on the existence of an
agent and a patient role both of which have a concrete non-linguistic referent. First,
the agent is responsible for the occurrence of the event and second, there has to be
a patient targeted by the action. The patient can be affected only if there is an agent
present. Typically, agent and patient roles are borne by different entities, but
reflexive readings of inherently transitive events are possible as well. The nature of
typical inherently transitive events implies a very high degree of agency (at least as
regards the features of intentionality and control). This means that these kinds of
action cannot be targeted at an entity indirectly and unintentionally, but the targeting
is always intentional and controlled. The agent is also aware of the consequences of
his actions. Hence, the degree of agency is usually very high and can vary only to
a limited degree. For example, an accidental instigation of these events is usually
excluded. This means that clauses like he painted the house accidentally are
(semantically) odd at best (cf. (414)-(422) below). We must adjust our actions due
to the goal we want to achieve. Since the occurrence of inherently transitive events
depends directly on the action of an agent, the event can proceed only as long as the
agent continues to act. The duration of the action of the agent corresponds directly
to that of the whole event. As a result, the agent is capable of controlling the degree
of affectedness, since the event affects the patient only as long as the agent is acting.
Causative events are underlyingly intransitive events that are caused by an
external agent. In their intransitive uses, they involve only a patient participant. The
underlying intransitive events and the corresponding causatives are primarily
distinguished on the basis of the presence of an agent in causatives. The patient is
equally affected in both cases. In typical cases, the introduction of an agent results
in a complete transitivization of an originally intransitive event. The introduction
of an agent also introduces the relevant agency features to the event. The most
salient feature of causative events is perhaps the obvious change-of-state in the
patient participant. However, the affectedness of patients involved in causative
events is in principle completely independent of external causation, since the
change-of-state can also occur spontaneously. Only if there is also an agent present
66

can we speak of transitive causative events. Since the underlying intransitive event
can occur without any involvement of an external agent, the agency in causative
events can vary greatly. This also enables indirect causation. The interval between
the initiation and the completion may be very long, which is not possible in
inherently transitive events. The agent can usually control the initiation part of
causatives, but may be unable to control its completion. For example, we may
initiate someone’s death, but we may be unable to prevent this person from dying
(unless we are trained doctors). Once initiated, the intransitive event may proceed
on its own. This is not to say, however, that the patient could control the change-of-
state it is experiencing in any way either. It is clearly a passive participant in
causative events. Consequently, the agent can control the duration of the event only
to a limited degree. The event usually proceeds as long as the patient has been
totally affected.
The third type of transitive event is presented by extended intransitive events.
Extended intransitives are similar to causatives in that they also have an underlying
intransitive reading. The underlying intransitive events have an existence of their
own, and their occurrence does not rely on the presence of two distinct entities. The
most significant difference between extended intransitives and causatives lies in the
nature of the participant introduced. In the case of extended intransitives an agentive
event is extended to include another participant. Since this produces a transitive
event, the introduced participant must be regarded as a patient. The inclusion of the
patient is usually purposeful. In causatives, the patient usually cannot volitionally
include another participant nor is it able to choose what entity initiates the event that
is targeted at itself. Only active participants in events are capable of intentionally
directing their actions at other participants. Hence, we restrict the use of the label
‘extended intransitives’ to the kinds of event referred to here. Extended intransitives
differ crucially from the other two event types discussed here because of the nature
of the event itself and the properties of the patient participant. Inherently transitive
events and causatives are both clearly single events with two distinct participants
one of which is an agent and the other a patient. The agent is responsible for the
active part of the event, whereas the patient is a passive participant that registers the
result of the event. This is the case also in some extended intransitives, such as ‘he
 67

spat on me’ (in distinction from ‘he spat’) and ‘he drove me there’ (as distinct from
‘he drove’). However, there are also extended intransitives that should rather be
described as consisting of two distinct events that are construed as one. For
example, in the event of ‘chasing’ there are two participants involved, both of which
are actively participating in the event (both of them are moving fast). The patient
cannot be construed as a mere passive participant, since it has some qualities usually
considered agentive. Consequently, active and passive features of the event are not
distributed ‘transitively’. Since extended intransitive events are primarily based on
agentive events that happen to be extended to other participants, they are usually
initiated purposefully and intentionally. However, the agency may vary more than
in the case of inherently transitive events that always imply a very high degree of
agency. For example, chasing someone or taking someone somewhere are usually
controlled and intentional events. Non-agentive readings of these events are rather
bizarre. On the other hand, spitting on someone can be intentional (as a sign of
hatred) or accidental. Furthermore, the degree of affectedness associated with
patients can also vary greatly in extended intransitives. Inherently transitive and
causative events usually result in a salient change-of-state in the patient participant
that registers the effects of the event (in the former cases the degree of completion
may vary). The degree of affectedness in extended intransitives varies from rather
saliently affected participants (he spat at me) to patients that are not in any salient
way affected by the events in which case there is no salient result (e.g. he chased
me). The group of extended intransitives is consequently somewhat more
heterogeneous than the other two event types as regards the degree of semantic (or
ontological) transitivity associated with them.
The threefold division illustrated above is primarily based on prototypical
instances of the different events. It is, however, important to note that the
conceptualization of an event as one of these types can be a default choice. A typical
example of this is provided by causatives in which both participants are inanimate
patients. We refer here to cases like ‘the tree fell on the car smashing it’. In this
case, a patientive intransitive event results in another one. Since transitive events
involve an agent (or an instigator in more general and neutral terms) and a patient,
we cannot but interpret the falling tree in the example cited as the ‘agent’. It is far
68

from being a typical agent, but is in this case interpreted as the more agentive entity
involved in the event. In somewhat similar way two running entities can be
construed as an instance of ‘chasing’. Both entities involved are acting volitionally
(cf. above), but the one not acting completely volitionally is viewed as the patient
in this case. The kinds of event referred to here are also somewhat less transitive,
because they involve two distinguishable events that happen to be construed as one.
To summarize. As briefly presented in this section, there are three differently
motivated transitive event types that are here labelled as inherently transitive,
causative and extended intransitive events. It is in order to say that these event types
have been analyzed only in light of couple of illustrative examples. Especiallym in
the case of extended intransitives there are many different subgroups, but a detailed
presentation of these is not relevant in the present section. As has been illustrated,
agency is the more relevant transitivity feature in the case of inherently transitive
and extended intransitive events. Of these two, high degree of agency is more
obvious in the case of inherently intransitive events, since they always involve
highly agentive instigators. These events can occur only if there is an agent present
that initiates the event in question. Causative events include an independent
patientive event whose occurrence per se is not agency-conditioned in any way. The
basic nature of these event types has clear consequences for the nature of transitivity
alternations applicable to a given event. As is semantically quite evident, the most
typical alternations are related to the affectedness of the patient, since high degree
of agency is conditioned by the event itself. Similarly, alternations applicable to
causative events are typically associated with agency, since the occurrence of the
underlying intransitive event does not rely on the participation of the agent and
furthermore, the patient is usually totally affected. Extended intransitives are the
most heterogeneous of these event types also as regards the nature of the relevant
transitivity alternations. Since there are two entities both of which have some
agentive features, changes in both the instigating and the affected participant are
equally possible. The effects of the differential motivation on the nature of
transitivity alternations will be discussed in more detail throughout the present
study.
 69

3.3. Aspects of structural transitivity

In this section, we will illustrate the concept of structural transitivity by

exemplifying the relevant features with concrete examples from a variety of
languages. This elaborates the definitions illustrated in 2.2. Since we will dwell on
the notion of structural transitivity later, only some illustrative examples will be
presented here. The goal is to show what aspects of argument marking and verb
morphology are relevant for the structural transitivity of clauses.
To begin with, we can safely state that there is nothing extraordinary in the
morphosyntactic form of transitivity markers. This means that the mechanisms
employed are typical linguistic elements, such as affixes (prefixes, infixes and
suffixes), adpositions (pre- and postpositions) or different kinds of particles. As
regards the marking of transitivity on verbs, different cross-reference affixes also
make a very significant contribution. Furthermore, inflectional and phonetic
elements (e.g., tone, stress, vowel length) are used in the transitivity marking. Also
reduplication of verb (stems) and incorporation of nouns are related to transitivity
in many languages (this is especially typical of Oceanic, especially Micronesian
languages, see e.g. Crowley 1982: 154). A couple of typical examples of transitivity
markers are given in (1)-(6), cf.

(1) thuku nuu-mi kulapuru-thi

dog lie-FUT blanket-LOC
‘The dog will lie on the blanket’
(2) thuku-yu nu-nti-mi kulapuru
dog-ERG lie-APPL-FUT blanket
‘The dog will lie on the blanket’(Kalkatungu, Blake 1987: 69) 21

21
The abbreviations used in the glosses are listed and explained at the beginning of the study. Some
abbreviations are not given, since we have not been able to find explanation for them. All relevant
abbreviations are, fortunately, listed. The use of different abbreviations for the same phenomenon
by different authors has been harmonized in order to avoid misinterpretations. Notice also that some
of the diacritics used in the original examples may have been omitted due to our inability to mark
them appropriately. This should not, however, have any consequences for the interpretation of
examples, since the possibly omitted diacritics are irrelevant for the purposes of the present study.
70

(3) naa hàrbi bàreewaa

1SG.PERF shoot(II) gazelle
‘I shot a gazelle’

(4) naa hàrbee shì à jikà

1SG.PERF shoot(II) 3SG.M.ACC in body
‘I shot him in the body’ (Hausa, Heide 1989: 28, 31)

(5) thuku-yu yuru-ø icai-na

dog-ERG man-ABS bite-PAST
‘The dog bit the man’ (Kalkatungu, Blake 1977: 17)
(6) laa pilapila-ø wanti-wanti õai-õu
he child-ABS follow-follow me-LOC
‘The kid keeps following me’ (Kalkatungu, Blake 1979: 33)

Functionally, transitivity markers are typically case markers, or in the case of verbs
different valence markers or cross-reference affixes. As regards the case marking,
the markers used in highly transitive clauses are usually so-called core
(functional/grammatical) cases, such as nominative/absolutive, accusative and
ergative. Other cases (like locative, dative, instrumental) are used in less transitive
clauses. These markers are typically semantic cases that have other functions as
well. When used in argument marking in less transitive clauses, the original
semantics of the case is usually bleached. For example, in (6) the locative marking
of the Patient is not related to any genuine locative semantics, whereas in (1) it is
used in this sense. Examples (1)-(2) and (5)-(6) are purposefully from Kalkatungu
for the purpose of showing that a single language can use different mechanisms in
the structural expression of transitivity. Verbal markers (as regards the marking of
transitivity) are usually related to marking of valence or to argument marking on the
verb. Typical examples include e.g. passive and antipassive morphemes and cross-
reference affixes.
The relevant features of structural transitivity include (in order of decreasing
significance) the number of explicitly expressed arguments, the (case-)marking of
arguments and the morphosyntax of the verb (including morphological marking (for
example, for voice) and cross-reference of arguments/participants). The number of
explicitly expressed arguments is naturally significant, since only clauses containing
 71

two (or more) arguments can be considered transitive. This is also the most directly
observable and self-explanatory criterion of transitivity. Following Blake (1977:16),
this kind of transitivity is here referred to as syntactic transitivity. Syntactic
transitivity is used here in a somewhat broader sense and it comprises all cases in
which two arguments are somehow explicitly expressed. This means that the
arguments do not have to nominal in order that we could consider a clause
syntactically transitive. This formulation is the typologically most applicable one,
because of which we take it as our starting point. Three different kinds of
syntactically transitive clauses are illustrated below, cf.

(7) mies tappo-i koira-n

man.NOM kill.PAST-3SG dog-ACC
‘The man killed the dog’ (Finnish)

(8) man-nen slår pojke-n

man-DEF hit.PRES boy-DEF
‘The man hits the boy’ (Swedish)

(9) a-li-ki-soma
3SG-PAST-OBJ-read
‘He read it’ (Swahili)

These are all regarded as instances of syntactically transitive constructions despite

their obvious morphological differences. The exemplified differences are due to
language structure, and it would not be justified to classify only one of these as
syntactically transitive. This approach would ignore morphology. Examples above
illustrate differential morphosyntactic mechanisms of referring to participants in
events. In Finnish, participants are not only referred to as independent nominal
arguments, but they are also separated from each other morphologically: the Agent
appears in the nominative, while the Patient is marked in the accusative. The
situation is somewhat simpler in Swedish, since the arguments are referred to
explicitly, but not distinguished morphologically.22 The seeming ambiguity is

22
Pronouns have an independent accusative form.
72

resolved by the order of arguments. The order is Agent-Patient (as in English).

Example (9) from Swahili differs from those from Finnish and Swedish in that the
participants in the event designated are referred to only by verb prefixes.23 Since the
number of the prefixes employed is two, the clause in question is here considered
transitive.
Syntactic transitivity is typically self-explanatory. This means that we do not
have to explain the syntactic transitivity by referring to any facet of events other than
transitive valence when explaining syntactic transitivity. Events that involve two
participants are (possibly) realized as syntactically transitive constructions.
However, there are also languages in which the mere number of arguments gives an
uncomfortable result. This is exemplified in (10) and (11) from Kammu (Svantesson
1983: 77):

(10) §ò § p c§ ká §
I eat fish
‘I eat fish’
(11) k c̀c tèn rntèn
he sit chair
‘He sits in the chair’

Example (10) exemplifies the typical transitive pattern of Kammu. Example (11),
on the other hand, poses a possible problem for our analysis of syntactic transitivity.
The event denoted is clearly intransitive, since only one of the participants can be
considered core. ‘Chair’ refers to a mere location in (11). It is naturally the case that
the semantics of ‘sitting’ requires a location, which means that (11) could also be
considered transitive. However, every event requires some location and location in
cases like (11) cannot be considered a core participant that would make a significant
contribution to semantics. In distinction from transitive events like that profiled in
(10), the location is more salient in (11), since there is only more core participant
and the location is hence more easily focussed on linguistically. Constructions like
(11) follow from the highly analytic character of Kammu. Kammu is an analytic

23
They could also appear as independent arguments.
 73

language without any morphological (or other) case-marking. On the basis of the
number of syntactic arguments alone, we can nothing do but classify (11) as
transitive. This is rather unfortunate, since semantics does not support our claim.
Here we can proceed in one of two ways. First, we can view syntactic transitivity as
a phenomenon that is totally independent of semantics. In this view cases like (11)
are instances of transitive clauses. The other possibility is that we specify the label
of syntactic transitivity somewhat by including some semantics. In this case, only
(10) exemplifies a syntactically transitive construction. In a strictly structural sense,
the former approach is more applicable. We simply have to explain the existence of
such ‘peculiar’ examples as (11) by the morphological nature of languages.
However, the latter does more justice to the notion of transitivity in general.
Analyzing constructions like (11) as transitive seems intuitively unjust. If we take
account of the nature of the participants (i.e. only core participants are included), we
can exclude clauses like (11) from the notion of syntactic transitivity (cf. section
4.5.).
In languages like Swedish, Kammu and English, (nominal) arguments of
transitive events are not morphologically distinguished. Consequently, the whole
concept of transitivity could be largely based on syntactic transitivity alone.
However, typologically mere syntactic transitivity gives too simple an idea of
transitivity. The inadequacies of syntactic transitivity as the only criterion for
transitivity were already briefly discussed above. A more typical and widely attested
example of such a criterion is provided by the differentialcase-marking of
arguments. Not only the number, but also the marking of arguments must be taken
account of when defining the transitivity of clauses. This is here (also in the spirit
of Blake (ibidem)) viewed as morphological transitivity. Examples are given in
(12)-(14) from Finnish in which the Patient may appear in different case forms, as
shown below, cf.

(12) mies sö-i leivä-n

man.NOM eat-PAST.3SG bread-ACC
‘The man ate the bread’
(13) mies katso-i poika-a
man.NOM look.at-PAST.3SG boy-PART
74

‘The man looked at the boy’

(14) mies suuttu-i poja-lle
man.NOM get.angry-PAST.3SG boy-ALL
‘The man got angry at the boy’

All these are syntactically transitive clause based on the number of explicitly
expressed arguments (defined as linguistic entities referring to participants in events
regardless of their (non-)core status). However, if we regard all these clauses as
transitive, we ignore significant morphological and semantic differences that are
integral parts of transitivity in Finnish. Differently from syntactic transitivity, we
must resort to non-linguistic cues when defining the transitivity of clauses. For
example, of the examples cited above, (12) is the only one usually considered
prototypically transitive, whereas (13) and (14) display lower degrees of transitivity.
We can justify this claim only by referring to the non-linguistic nature of the events
described. (12) is the only clause that profiles an ontologically transitive event. The
ontological (and consequently, morphological) transitivity of (13) and (14) is
considerably lower. There is nothing in the mere morphological form of arguments
that would enable us to regard certain clauses as more transitive than others. As a
result, a linguist not familiar with the transitivity in Finnish could consider (13) as
the transitive archetype of Finnish. Only when s/he makes him/herself familiar with
the structure of Finnish, does s/he learn that this claim is not justified, since it is in
clear contradiction with the semantics of events. By simply stating that (12)
illustrates the basic transitive structure of Finnish, we ignore the semantic
motivation of the marking.
The morphological transitivity is not restricted to cases in which cases are
marked morphologically on the arguments. It is also possible that case is marked,
for example, with adpositions, as in Japanese24, cf.

(15) Matti ga otoko o korosh-ita

PN NOM man ACC kill-PAST
‘Matti killed the man’

24
All Japanese examples used in the present study are courtesy of Nobufumi Inaba.
 75

(16) kare ga tokyo ni ikimas-u

he NOM Tokyo DAT go-PRES
‘He is going to Tokyo’

In (15), the Agent appears in the nominative (marked with ga), whereas the Patient
is in the accusative case (o). We can justify the high transitivity of (15) with the
ontology of the denoted event. Furthermore, as illustrated in (16) the non-subject
argument of less transitive clauses often appears in a case other than the accusative.
Example (16) must be classified as less transitive similarly to (13) and (14). In both
these languages, the morphological transitivity is relevant. It just happens to be
marked differently (which, of course, is due to the structure of these languages).
Morphological transitivity is usually a far better indicator of ‘true’ transitivity
than the mere number of arguments. By taking account of case marking, we are able
to make finer distinctions between two clauses. Morphological transitivity usually
reflects the underlying semantics somehow. The structure employed in the
expression of highly transitive events is viewed as highly transitive, while those
marked differently are regarded as less transitive. However, morphological marking
of arguments does not always enable us to classify clauses correctly according to
their transitivity, but there are always ‘quirks’ First, case-marking correlates only
roughly (at best) with the semantic transitivity of events, cf.

(17) mies näk-i talo-n

man.NOM see-PAST.3SG house-ACC
‘The man saw the house’
(18) mies lö-i poika-a
man.NOM hit-PAST.3SG boy-PART
‘The man hit the boy’ (Finnish)

In (17), the Patient appears in the accusative that is the case of typical Patients in
Finnish (among many other languages). In (18), the Patient appears in the partitive
that is a general indicator of decreased transitivity (it marks, for example,
imperfective aspect). On the basis of morphological transitivity, we could classify
(17) as the more transitive of these two clauses, even though the ontological
transitivity of (18) is higher.
76

Some kind of morphological (non-basic) marking may be typical of all non-

subject arguments, in which case we are dealing with a similar case as that
illustrated with examples from Kammu (see above). Examples of this kind of
morphological case-marking are attested, for example, in Creek (Martin 2000: 379):

(19) cá "ni ‘taló "fa-n hî"c-is

John town-OBL see:FGR-INDIC
‘John sees the town’
(20) cá "ni ‘talo "fa-n ay-áha "n-ís
John town-OBL go:SG-FUT:LGR-INDIC
‘John is going to town’

The ontological transitivity of both (19) and (20) is rather low. However, (19) can
be regarded as more transitive, since it refers to an event inherently transitive in
valence. The transitive marking can consequently be said to be justified in this case.
The event described in (20), on the other hand, is clearly intransitive, which makes
the transitive marking totally unmotivated. In languages like Creek, a kind of
nominative-accusative (unmarked-oblique) marking is not a sufficient criterion for
(morphological) transitivity, since this structure can be employed in the linguistic
marking of completely intransitive events as well.
So far, we have dwelled on the number and marking of arguments as relevant
features of linguistic transitivity. Moreover, the verb morphology and the cross-
reference of the verb also contribute to structural transitivity. These features are here
classified as parts of morphological transitivity, since they are primarily
morphologically marked and since we usually must resort to features other than
mere syntactic transitivity when defining the significance of different verbal markers
for transitivity. There are no such structural properties of verbs that would be
directly and universally associated with certain features of high or low transitivity,
but these are highly language-specific features. An aspect of transitivity that is
marked on the verb in one language may be totally irrelevant in others.
Consequently, it is impossible to say whether some kind of verbal marking in a
specific language is related to high or low transitivity. Different features of the verb
related to the marking of transitivity in different languages are illustrated in what
 77

follows. We begin by examining morphological features related to voice/diathesis

and proceed to cross-reference.
Basic clauses are usually basic also as regards the marking of the predicate
verb. The verb typically appears in a zero form. The number or the marking of
arguments does not have any direct effect on the marking of the verb. Examples are
illustrated in (21)-(23) from German:

(21) er schläg-t den junge-n

he.NOM hit.PRES-3SG ART.ACC boy-ACC
‘He is hitting the boy’
(22) er hilf-t der frau
he.NOM help.PRES-3SG ART.DAT woman
‘He is helping the woman’
(23) er flieg-t nach münchen
he.NOM fly.PRES-3SG PREP Munich
‘He is flying to Munich’

Examples illustrated above differ crucially from each other in their semantic and
also morphosyntactic transitivity. This does not, however, result in any changes in
the morphology (or cross-reference) of the predicate verb, but the verb appears in
every case in a basic (active) form. The number and marking of the arguments is
determined by the semantics associated with the verb. The verb itself remains
unaffected.
The majority of languages are like German in that they do not mark verbs of
basic clauses explicitly for transitivity. However, examples of languages in which
the transitive valence is marked also in basic cases are not that hard to find.
Illustrative examples are found, for example, in Salish languages, as shown below,
cf.

(24) g wcNq ’k w-cs-w c-s

FUT N bite-TR-2SG.OBJ-3.TR.SUBJ
‘He will bite you’ (Tillamook, Kroeber 1999: 24)
(25) niN lcm-c-›-él- cm §c N|cN s |éni§
AUX N see-LV-TR-1SG.OBJ-AD OBL N ART N woman
‘I was seen by the woman’ (Salish, Kroeber 1999: 26)
78

(26) n N sát-šit-umi
1SG.SUBJ.CL N give-DITR-2SG.OBJ
‘I give it to you’ (Salish, Kroeber 1999: 30)

Examples (24) and (25) illustrate basic transitive patterns, whereas (26) exemplifies
a ditransitive clause pattern of Salish. In (24) the verb ‘bite’ is followed by a verbal
affix that is glossed as a transitive marker. ‘Biting’ is an inherently transitive event,
since it always requires an agent who is biting someone or something.
Consequently, the transitive marking of the verb is semantically very redundant.
Example (25) exemplifies agent demotion. In this case, the expected transitive
marker is replaced with an agent demotion marker (Kroeber 1999: 25). Despite the
decreased morphological transitivity of the given construction the transitive marker
is retained. Therefore, it seems that the function of the marker is to mark the
transitive valence of clauses, but that of events. Example (26) shows that also
ditransitive verbs are marked according to their transitive valence.
The verb usually appears in some kind of basic form in basic clauses. This
means that the predicate verb is not marked explicitly for transitivity in basic clauses
(it may, of course, be marked for cross-reference or the like). The zero marking of
the verb is usually restricted to cases in which the transitive valence agrees with the
number of arguments expressed. Some kind of explicit (i.e. non-basic)
morphological marking of the verb often signals changes in transitivity. Typical
examples of this are changes in the valence (in the traditional, syntactic sense) of the
verb, cf.

(27) der mann wird geschlagen

ART.NOM man become.PRES.3SG hit.PARTIC
‘The man is hit’ (German)

(28) kala syö-dään

fish.NOM eat-PASS.PRES
‘The fish is eaten’ (Finnish)

(29) otoko wa koros-are-ta

man TOP kill-PASS-PAST
‘The man was killed’ (Japanese)
 79

Examples (27)-(29) all describe transitive events. However, (27)-(29) illustrate

clauses in which the structural transitivity and transitive valence disagree, since only
one of the two participants is explicitly referred to. This change in valence of the
verb is explicitly marked on the verb. A verb that is marked as passive cannot take
two core arguments in German, Finnish or Japanese. If we want to express the
Agent explicitly, it must appear in an oblique form. The kind of marking
exemplified here is very economical, since only cases that deviate somehow from
the basic transitive valence of the verb are marked explicitly.
Examples (27)-(29) exemplify cases in which a reduced valence of the verb
is marked explicitly. It is, of course, possible that the function of the marking is to
signal an increase in the typical valence. An example is given below (Turkish,
Comrie 1989: 175f):

(30) Hasan öl-dü

Hasan.NOM die-PAST
‘Hasan died’
(31) Ali Hasan-i öl-dür-dü
Ali Hasan-ACC die-CAUS-PAST
‘Ali killed Hasan’

Example (30) illustrates the typical intransitive clause of Turkish, whereas (31) is
a causative derived from (30). The valence of the verb has increased by one, which
is signalled on the verb by the causative affix -dür. Consequently, we know that the
given verb is transitive and can consequently take two core arguments. Also in this
case, we can infer the number of participants partaking in the event described from
the morphological form of the verb. This is not the case with every verb, but this is
in general restricted to explicitly marked verbs. The exemplified case from Turkish
differs from those illustrated in (27)-(29) in that we are dealing with a change in the
transitive valence of the event, whereas in (27)-(29), the change affects only the
syntactic valence of verbs (the event itself remained the same). In (31), the marking
is semantically motivated.
In (27)-(31), changes in the valence of verbs are marked explicitly. However,
80

as already illustrated in light of German (see (21)-(23)) the explicit marking of

valence is typically restricted to deviant cases. In basic clauses, we have to resort to
verb semantics when defining the transitive valence. The verb morphology is
consequently only a partial indicator of transitivity at best. Furthermore, in a number
of languages, many verbs are neutral as regards the valence, but the verb can appear
in more than one case frames. An example is illustrated below, cf.

(32) get’e xa-na

pot break-AOR
‘The pot broke’
(33) zamira.di get’e xa-na
Zamira.ERG pot break-AOR
‘Zamira broke the pot’ (Lezgian, Haspelmath 1993: 289)

Examples from Lezgian exemplify typical intransitive and transitive clauses,

respectively. As can be seen, the verb xana can appear in both an intransitive and
a transitive frame without any change in its morphological form.25 In cases like
these, the correlation of morphological marking with any kind of transitivity is even
more indirect than in languages like German, since one and the same verb can
appear in multiple case frames.
Above, we have exemplified cases in which the marking on the verb has
signalled changes in the valence (either semantic or syntactic) of the verb. It is also
possible that morphological changes in the verb are related to individual facets of
transitivity. An example is given below, cf.

(34) õa-s dkaryol bcaq-pa-yin

I-ERG cup break-PERF-VOL
‘I broke the cup purposefully’
(35) õa-s dkaryol bcaq-so õ
I-ERG cup break-PERF.INVOL
‘I broke the cup accidentally’ (Lhasa Tibetan, DeLancey 1984: 132)

25
Haspelmath shows in light of imperatives, scope of negation and ‘Involuntary Agent constructions’
that (33) is a labile verb and not a construction with a missing ergative.
 81

In this case, the number of participants and arguments remains the same despite the
morphological changes in the verb. In Lhasa Tibetan, the semantic difference
between the events denoted is found in agency. Structurally, this difference is
highlighted by choosing the appropriate ‘volitionality affix’.26 This change produces
less radical changes than those in (30) and (31), for example.
Above, we have briefly illustrated the relation of transitivity and verbal
morphology in light of examples from a couple of languages. The other significant
verbal feature associated with the expression and marking of transitivity is the cross-
reference of argument(s) on the verb. Cross-reference is more directly related to the
transitivity of events than many aspects of verb morphology and it can also be
regarded as a kind of argument marking.
There are two different kinds of cross-reference that stand in a differential
relation to the marking of transitivity. The first cross-reference type refers to the
kind attested in, for example, Finnish and German. In these cases, the verb agrees
with the morphologically unmarked argument (that in German and Finnish appears
in the nominative case) irrespective of its semantic role or the transitivity of clauses.
It is not possible to cross-reference both arguments in these languages. Hence, this
kind of cross-reference does not relate in any direct way to transitivity, but we are
rather dealing with an obligatory subject agreement that is conditioned by the status
of arguments.
The other type of cross-reference can be regarded as more important as regards
the marking of transitivity. In this type, the arguments appear on the verb as affixes.
This kind of cross-reference can be seen as a sub-type of syntactic marking of
arguments. The only difference between these two is that in one the arguments are
expressed as independent arguments, whereas in the other they are attached to the
verb. There are two subtypes, both of which are illustrated below, cf.

(36) aywa-n
go-3SG
‘He goes’

26
According to Scott DeLancey (p.c.) examples like (34) and (35) are regarded as transitive and
intransitive version of ‘break’ in standard grammars of Tibetan.
82

(37) maqa-ma-n
hit-1SG-3SG
‘He hit me’ (Quechua, Payne 1997: 136)

(38) Péter olvas egy könyv-et

P. read/3SG(SUBJ) a book-ACC
‘Peter reads a book’
(39) Péter olvassa a könyv-et
P. read/3SG(OBJ) ART book-ACC
‘Peter reads the book’ (Hungarian, Lazard 1998: 169)

(40) agna-m ukini- ±a atkuja-±

woman-ERG sew-3SGA.3SGO dress-ABS
‘The woman is sewing the dress’
(41) agna-± ukini- ±u ± (atkujagmyn)
woman-ABS sew-3SG (dress.INSTR)
‘The woman is sewing (a dress)’ (Asiatic Eskimo, Kazenin 1994: 146, cited from
Menovš ikov 1967 and Vaxtin 1981)

In Quechua, the cross-reference is very clearly another form of argument marking.

Example (36) illustrates the typical intransitive clause of Quechua. The event at
issue involves only one participant that is cross-referenced on the verb. The number
of affixes is two in (37), since the event in question is transitive. The number (and
form) of affixes is conditioned directly by the transitive valence of the verb. The
examples from Hungarian are somewhat more interesting. In Hungarian (as in some
other Finno-Ugric languages, including Mordvinian (cf. (314)-(315)), but excluding
Finnish), there are two different kinds of agreement that are labelled as subjective
and objective conjugations. The difference between these is that the subjective
conjugation is with the subject only, whereas in the objective conjugation both
arguments are cross-referenced. In distinction from languages like Quechua, the
cross-reference is not determined by the number of arguments alone, but their
morphosyntactic nature also makes a contribution. The cross-reference in Asiatic
Eskimo is somewhat similar to that in Hungarian, since only core arguments can be
cross-referenced on the verb. Also in Asiatic Eskimo, the number of arguments per
se is irrelevant. In Hungarian and Asiatic Eskimo, the cross-reference is clearly an
 83

additional mechanism of transitivity. In both languages, the transitive cross-

reference is restricted to (structurally) highly transitive clauses.
Examples (36)-(41) illustrate a kind of syntactic marking that is only indirectly
conditioned by the semantic transitivity of events. In these kinds of case, the
semantics of the verb is primarily responsible for the interpretation of transitivity.
The agreement affixes only make a minor contribution to the interpreted transitivity.
The other possibility is that the verb affixes are more directly related to the
expression of transitivity. Examples are given below:

(42) hattak-at oho:yoh(-a) ø-ø-pisa-h

man-SUBJ woman(-OBL) 3AGT-3PAT-see-PRES
‘Man sees woman’
(43) hattak-at oho:yoh(-a) ø-i-hiyiya-h
man-SUBJ woman(-OBL) 3AGT-3DAT-stand-PRES
‘Man waits for woman’
(44) hattak-at oho:yoh(-a) i-ø-nokšo:pa-h
man-SUBJ woman(-OBL) 3DAT-3PAT-be afraid-PRES
‘Man is afraid of the woman’ (Choctaw, Heath 1977: 207)

In these examples, the participants are referred to by both independent arguments

and (dependent) verb affixes. The contribution of these mechanisms to the overall
transitivity of clauses is very different. The two mechanisms are further completely
independent of each other. The primary function of the case marking on independent
arguments can be said to be a disambiguating one: based on this marking we know
which participant bears which role. The verb affixes, on the other hand, are related
to the inventory of semantic roles. They explicate which roles are present. The
affixes are in these cases more directly related to the expression of semantic
transitivity. If transitivity of a clause is rather low, affixes referring to typical agent
and patient, can no longer be used. Only the inventory of affixes reflects the
semantic transitivity of events, since the marking of independent arguments is
constant in all these examples.
Since the features of verbs associated with the marking of transitivity are best
regarded as components of morphological transitivity (in the sense the label is used
here), we must resort to the semantics when defining their effect on transitivity of
84

clauses. Furthermore, the basic structure of languages (including, for example, the
basic argument marking pattern) influences the marking of transitivity on the verb.
The latter is especially important, as regards the nature of transitivity alternations.
For example, some transitivity feature may be marked explicitly in one language,
whereas the opposite is marked in another one. The nature and number of verbal
affixes employed are also very language-specific phenomena. There is no single
affix that would be universally used for expressing, for example, the lower degree
of affectedness of the patient participant.

3.4. Transitivity of meaning and form

In the previous section, we illustrated some relevant aspects of structural transitivity

in more detail. Below we will discuss the underlying motivation (focussing on the
number and form of arguments). The notions at issue are here referred to as
transitivity of meaning and form. The former refers to cases in which the number of
arguments is determined by the semantics of events or verbs. The latter, on the other
hand, comprises cases in which the morphosyntactic features of clauses (typically
morphology of the verb) have consequences for structural transitivity.
Transitivity of meaning refers here to the number and marking of arguments
determined by the non-linguistic nature of events. This means that the number and
marking of arguments reflect the inherent transitivity of events, for example, in light
of the number of participants or their semantic roles.27 The morphology of verbs is
as such irrelevant, since the semantics of verbs tells us what kind of event is referred
to. Morphologically identical verbs can denote radically different events.
Transitivity of meaning is here regarded primarily as a feature of events, not the
semantics of verbs. This distinction is crucial, since there are cases in which the
semantics of verbs and that of events disagree without any morphological marking
that would motivate the differential behaviour. These kinds of case are intermediate
between instances of transitivity of meaning and form. In the present study they are

27
It might be in order to explicitly state that occasional/optional omission of arguments is excluded
from the description here. This means that cases in which transitive verbs allow the expression of two
arguments are taken account of irrespective of whether either or both of them can be omitted.
 85

understood as a subtype of transitivity of form, since the basic meaning of the event
is not responsible for the marking, but rather the linguistic reference to the event.
Transitivity of form also comprises cases in which the marking is due to formal
properties of verbs (such as reflexive, passive or antipassive marking of the verb).
This means that the morphological form of the verb requires certain kinds of
morphosyntactic marking from its arguments. This form may also be semantically
motivated (as for example in the case of (anti)causative), but what is important for
our purposes is that the marking clearly is different from that required by an
unmarked predicate. The marking may also deviate from the form conditioned by
the semantics in basic cases. Both underlying motivations of transitivity are
discussed below.
Relevant semantic properties of events that condition the transitivity of clauses
include the number of participants, their semantic roles and the overall transitivity
associated with the event (including kinesis etc.). In typical cases, the number of
participants and arguments correlate perfectly with each other. Examples include the
following, cf.

(45) mies kuol-i

man.NOM die-PAST.3SG
‘The man died’
(46) mies tappo-i karhu-n
man.NOM kill.PAST-3SG bear-ACC
‘The man killed the bear’ (Finnish)

In (45) and (46), the number of explicitly expressed arguments is directly

determined by the transitive valence of the verb. The event profiled in (45) is
intransitive, and the number of arguments expressed one. In similar vein, the
number of arguments in (46) is two, since the event is transitive.
Above, we illustrated cases in which the number of explicitly expressed
arguments correlates directly with the number of participants involved. In addition
to the number of arguments, the inherent semantics of verbs is often also responsible
for argument marking. This is the case also in (46), where the Agent appears in the
nominative case, whereas the Patient appears in the accusative. This is the typical
86

transitive pattern of Finnish. That the marking in (46) is truly motivated by the high
transitivity of events can be illustrated by comparing it with the marking of clauses
that describe events that are clearly less transitive than that in (46), cf.

(47) mies ajattel-i nais-ta

man.NOM think-3SG.PAST woman-PART
‘The man was thinking about the woman’
(48) mies rakastu-i naise-en
man.NOM fall.in.love-3SG.PAST woman-ILL
‘The man fell in love with the woman’ (Finnish)

Neither (47) nor (48) denotes a transitive event. The lower degree of transitivity
manifests itself in the argument marking, i.e. changes in the marking coincide with
changes in meaning. The Agent remains in the nominative, but the marking of the
Patient is manipulated. It appears either in the partitive or in the illative case. The
marking of the Patient is motivated by the semantics of events, since the Patient
appears in the accusative case only if the event described is highly transitive.
Conversely, the Patient never appears in the illative case if the event described is
highly transitive.
Interesting examples of cases in which the marking of transitivity is clearly
conditioned by the non-linguistic transitivity are provided by the next examples, cf.

(49) àng wà-apm-ì

3SG say-TMdys-3.INTR.PAST
‘He said something’
(50) à:ng-í wà-apm-à
3SG-AGT do-TMdys-3.TR.PAST
‘He did something’ (Dulong/Rawang, LaPolla 2000: 287)

(51) sa-dííy nurutí-íva

3SG-see alligator-DAT
“He saw an alligator’
(52) sa-dííy nurutú-Ø
3SG-kill alligator-ACC
‘He killed an alligator’ (Yagua, Payne 1997: 101)
 87

Examples (49)-(52) illustrate differential behaviour of homophonous verbs. The

interpretation of clauses depends in this case crucially on the morphological marking
of arguments, since two clearly different interpretations are possible. In examples
like (45)-(48), the lexical form of the verb makes a significant contribution to the
interpretation of transitivity, since the relevant facets of transitivity are lexicalized
components of the verbs in question. The marking of arguments simply emphasizes
the transitivity features inherent in the verb. In (49)-(52), on the other hand, we can
infer the intended meaning of the two possible ones only from the marking of the
Agent (Dulong/Rawang) or the Patient (Yagua). Hence, in these kinds of case the
structural transitivity of clauses is clearly motivated by the semantics of events
denoted.
In (49)-(52) from Dulong/Rawang and Yagua, homophonous verbs appear in
different case frames depending on the ontological transitivity of events. In both
languages, the former example describes an event displaying a rather low degree of
transitivity, whereas the latter cases designate typical transitive events. In (49)-(52),
the number of arguments (along, naturally, with the number of participants
involved) is the same, but the marking varies. The other possibility is that the
number of arguments expressed varies. A couple of examples of this are illustrated
in (53)-(57), cf.

(53) y-ttcu wqqzin (aysum)

3MS-eat dog.CST (meat)
‘The dog ate (the meat)’
(54) y-ttcu yghzer tafunast
3MS-eat.PERF river.CST cow
‘The river drowned the cow’
(55) y-ttcu yghzer
3MS-eat.PERF river.CST
‘The river ate (something)’ (*the river drowned) (Berber, Guerssel 1986: 36ff)

(56) mies pol-tt-aa ??(talo-n)

man.NOM burn.TRANS-CAUS-3SG.PRES ??(house-ACC)
‘The man is burning a house’ ‘??The man is burning (something)’
88

(57) mies pol-tt-aa (tupakka-a)

man.NOM burn.TRANS-CAUS-3SG.PRES (cigarettes-PART)
‘The man is smoking (cigarettes)’ (Finnish)

These examples differ from those illustrated above in that the basic semantics of
verbs remains more or less the same (but not identical). Only the nature of
participants varies (which also affects the semantics of verbs). This has clear
consequences for the expression of arguments. In Berber, the verb ttcu has different
meanings. The basic meaning is ‘eat’. In this use, the verb allows the Patient to be
freely eliminated, as in (53). Example (54) illustrates a metaphorical use of ttcu. As
can be seen in (55), the Patient omission is restricted to the basic meaning. Example
(55) is infelicitous, since it implies that river actually consumed some food, as
humans do. In this case the semantic nature of the instigator (or cause) participant
is responsible for the ill-formedness. The case is converse in Finnish in two ways.
Clause (56) describes the more basic meaning of the verb polttaa that is burning
things. In this use, the Patient is rarely omitted. The reason for this can be seen in
(57). If the Patient is deleted, the default meaning is ‘smoke’, as in (57). In both
cases, we are dealing with some kind of burning, but only one of these allows a free
omission of the Patient argument. Also in (53)-(57), the form of the verb is neutral
as regards the transitivity (transitive valence) of clauses, but the expression or
omission follows from the semantics of participants.
In the examples presented so far, the number and marking of arguments has
reflected the ontological nature of events denoted. It is also possible that the
marking is only indirectly motivated by the inherent (basic) semantics of events, but
the overall semantic transitivity makes a more significant contribution to the
marking. Examples are illustrated below, cf.

(58) mies jo-i mehu-n

man.NOM drink.PAST-3SG juice-ACC
‘The man drank the juice up’
(59) mies jo-i mehu-a
man.NOM drink.PAST-3SG juice-PART
‘The man drank (some) juice’ (Finnish)
 89

(60) cy-n c teb cl-d c the õõi

28
I-ERG table-LOC touched
‘I touched the table (volitionally)’
(61) cy teb cl-d c the õõi
I table-LOC touched
‘I touched the table (involuntarily)’ (Manipuri, Bhat & Ningomba 1997: 104)

At first sight, the examples illustrated above seem identical to those from
Dulong/Rawang and Yagua, since we are dealing with formally identical verbs
appearing in multiple case frames. What these examples have in common is that in
both cases the marking is clearly motivated by the semantics of events and not by
the form of the verb. In Dulong/Rawang and Yagua, the semantic differences
between events were very significant, whereas in (58)-(61) the events in question
differ from each other in a less significant way. Consequently, examples (58)-(61)
show that not only the inherent semantics, but also minor aspects are important. In
Finnish, the marking is conditioned by the degree of affectedness associated with
the patient. The Patient appears in the accusative, if it is regarded as totally affected.
The partitive marking correlates with imperfective aspect. In Manipuri, the relevant
feature is agency. Both profiled events are transitive and involve two participants
one of which is responsible for the event, whereas the other is targeted by it.
However, in (61) the agent is intending the action to take place, which results in a
zero marking of the Agent. In both cases two possible interpretations inhere in the
semantics of the verb and the marking is situationally determined. Hence, the
marking is not conditioned solely by the inherent nature of events, but we must take
other factors into account.
In cases above, the semantics of events more or less coincides with the
semantics of verbs. In other words, the semantic features associated with events and
participants are lexicalized parts of verbs. It is also possible that the verb semantics
and the semantics of events disagree. In these cases, transitivity is determined more
directly by the form of the verb. For example, the number of participants partaking
in the event and that allowed by the verb may be different, cf.

28
W e have replaced the label ‘nominative’ with ‘ergative’ in all examples from Manipuri.
90

(62) žek’-~ biš wa r-ac’-xo

man-ERG food.ABS IV-eat-PRES
‘The man eats the food’
(63) ¨ali Ø-iš-xo
Ali.ABS I-eat-PRES
‘Ali eats’ (Tsez, Comrie 2000: 363, 366)

(64) er ha-t (brot) gegessen

he.NOM have-3SG.PRES (bread) eat.PARTIC
‘He has eaten (bread)’
(65) er ha-t *brot gespeist
he.NOM have-3SG.PRES *bread dine.PARTIC
‘He has dined (*bread)’ (German)

Examples illustrated above exemplify intermediate cases between cases in which the
transitivity is directly conditioned by meaning and purely structural cases. In these
cases, somewhat untypical semantics of verbs is responsible for the disagreement
of structural and semantic transitivity illustrated above. Morphologically, verbs in
(63) and (65) are identical to typical verbs. Examples (62) and (64) exemplify
typical cases, since the event denoted is transitive and the clauses have two
arguments. Events denoted in (63) and (65) are more or less the same, but the
constructions employed are obligatorily intransitive. Neither (63) nor (65) allows the
Patient to be expressed overtly, even if it is an integral part of the given event. The
number of participants and arguments disagree in (63) and (65). Hence, the number
of arguments is a lexicalized part of the verb.
In (63) and (65), the reduced structural transitivity is not marked on the verb.
Opposite cases are frequent. This means that the number of arguments is motivated
by the form of the verb and not solely by the semantics of events or verbs. Typical
instances of this include the following:

(66) nainen tappo-i miehe-n

woman.NOM kill.PAST-3SG man-ACC
‘The woman killed the man’
(67) mies tape-tt-iin ??naisen toimesta
man.NOM kill-PAST-PASS by the woman
 91

‘The man was killed (by the woman)’ (Finnish)

(68) o 8u-l b cx koše

that/OBL-ERG grass/5 mow.PRES
‘Father mows the grass’
(69) o 8u koše-laa
that/1 mow-ANTIP.PRES
‘Father is mowing’ (Hunzib, Van den Berg 1995: 110)

(70) ùbúr á-p āār`

Ubur CMPL-jump
‘Ubur jumped’
(71) dháag ] á-p ārr`-ì ùbúrr-ì
woman CMPL-jump.BEN-SUF Ubur-ERG
‘Ubur jumped for the woman’ (Päri, Andersen 1988: 304)

Events denoted in (66)-(69) are all transitive. However, both participants explicitly
referred to only in two cases. Due to the disagreement of structural and semantic
valence, (67) and (69) are similar to (63) and (65). However, in (67) and (69), this
is morphologically signalled: (67) and (69) exemplify derived intransitives of
Finnish and Hunzib. The events profiled by these syntactically intransitive
constructions are transitive in valence. The structural intransitivity is not
semantically motivated, but it is due to the syntactic requirements of the predicate
verb. In (67) and (69), the verb has been morphologically de-transitivized (for a
more detailed analysis of de-transitivization, see 5.3.1.1.2.), which implies a
reduction in the number of arguments. In Finnish and Hunzib, the de-transitivization
is complete, since only one of the two participants can be explicitly referred to. In
(67) and (69) the number of arguments is lower than that of participants in the given
event. The opposite of this is exemplified in (71). Both (70) and (71) describe
intransitive events with one core participant. However, (71) involves two arguments
and the structure employed is transitive. Despite the semantic intransitivity of the
event in question, both arguments are syntactically obligatory in (71). Consequently,
(71) exemplifies a case in which a syntactically transitive construction is employed
in the description of an intransitive event. As in (67) and (69), the overt expression
and the marking are primarily conditioned by the morphological form of the
92

predicate verb in (71).

In examples above, the number of core participants and arguments obligatorily
disagree. This is due either to the peculiar semantics of the verb, as in (63) and (65)
or to the morphological marking of the verb, as in (67), (69) and (71). Moreover, the
verb morphology may also impinge on the argument marking only. This means that
the arguments are marked differently on the basis of verb morphology, as in
examples below, cf.

(72) bama-lu gurra: du:-ny

man-ERG dog.O hit-PAST
‘The man hit the dog’
(73) bama gurra:-nda du:-yi-ny
man.S dog-DAT hit-INTR-PAST
‘The man hit the dog’ (Djabugay, Patz 1991: 299)

(74) ich schlug ihn

I.NOM hit.PAST.1/3SG he.ACC
‘I hit him’
(75) er wurd-e von mir geschlagen
he.NOM become.PAST-1/3SG ‘by’ I.DAT hit.PARTIC
‘He was hit by me’ (German)

Examples (72)-(75) involve two explicitly expressed arguments. (72) and (74)
represent basic transitive clauses of Djabugay and German, whereas (73) and (75)
exemplify derived intransitives. In distinction from (67) and (69), the number of
arguments is not necessarily affected, but the changes are restricted to the
morphological marking and status of arguments. This also follows from
requirements of the verb. Since the explicitly marked argument of the basic clause
(Agent and Patient, respectively) has been promoted, the primary argument of the
basic clause has been demoted to an oblique status. The number of obligatory core
arguments has decreased.
4. MARKING OF (HIGH) TRANSITIVITY

4.1. Preliminaries

For the purpose of studying the typology of transitivity alternations, we also have
to define the concept ‘basic transitive clause’. This is the goal of this chapter. The
nature and the motivation of transitivity marking are both important in this respect
and will hence be illustrated and discussed thoroughly. We will examine what
aspects of transitivity determine the transitive marking of clauses. As has been
pointed out, transitivity is both a semantic and a structural phenomenon and both of
these aspects may contribute to transitivity. In the present chapter, we will dwell on
the motivation of transitive marking from different perspectives and propose a
typology of the basic transitive clause. The latter means that we will illustrate the
mechanisms employed in the marking of high transitivity.
Before proceeding to the topic of this chapter, we first have to define the
relevant notions. Linguistic transitivity is here understood to cover the linguistic
expression of any transitivity facet (with all their different degrees) discussed in 3.2.
Transitivity markers are here understood as morphosyntactic elements that are
employed in the linguistic marking of (different degrees of) transitivity. The explicit
marking may relate to high (or increased) or low (or decreased) transitivity. In order
to be classified as genuine transitivity markers, the markers under study must be
directly related to the expression of transitivity in general or to some specific aspects
of it. This function does not have to be the primary one (see 4.4.). The latter is
important, since the use of basic transitive clauses is not necessarily conditioned by
high transitivity. The number of participants referred to is here taken for granted, i.e.
it is not regarded as a feature of transitivity marking, but rather as a prerequisite.
Relevant aspects include the marking of arguments and the morphology of the verb.
One of the chief properties, as regards different kinds of transitivity marking,
is the distinction between semantically and structurally motivated marking types.
This means that the primary function of the marking may be either to emphasize the
structural transitivity of clauses or it may be primarily related to the expression of
certain aspects of transitivity. Typical examples of the former are given in (76)-(78),
whereas (79)-(82) exemplify the latter kind of marking, cf.
(76) thaa aara peba-ng ngatha
door that.ABS wind-ERG shut
‘The wind shut the door’ (Kugu Nganhcara, Smith & Johnson 2000: 390)

(77) mies tappo-i karhu-n

man.NOM kill.PAST-3SG bear-ACC
‘The man killed the bear’
(78) mies näk-i karhu-n
man.NOM see-3SG.PAST bear-ACC
‘The man saw the bear’ (Finnish)

(79) son Xm q Xjje ne-nt•he §

Sonam tiger PFV.TR-kill.PFV
‘Sonam killed the tiger’
(80) q Xjje-k c son Xm ne-s c-set
tiger-ERG Sonam PFV.TR-CAUS-die.PFV
‘The tiger killed the Sonam’ (Caodeng rGyalrong, Sun 1998: 130f)

(81) hasan öküz-ü ald x

Hasan.NOM ox-ACC buy.PAST.3SG
‘H. bought the ox’
(82) hasan bir öküz ald x
Hasan.NOM an ox buy.PAST.3SG
‘H. bought an ox (non-specific)’ (Turkish, Comrie 1989: 132)

In (76)-(78), the primary motivation of transitivity is structural. This means that the
syntactic transitivity determines the basicness of clauses. The transitive marking
appears by default in case two arguments are expressed. The examples from Finnish
illustrate two different kinds of structurally motivated marking. In (77), the marking
is considered structural, since the accusative marking of the Patient is responsible
for the expression of transitivity as a whole. In (78), the marking is considered
primarily structural, since the clause is typically transitive despite the considerably
lower degree of semantic transitivity of the event denoted. In Finnish, the primarily
structural motivation is also reflected in the fact that core arguments appear in the
accusative only if two arguments are expressed. The motivation is clearly different
in (79)-(82). At first sight, (79) and (81) seem like a typical structurally motivated
ergative and accusative clauses, respectively. However, as seen in (80), the ergative
 95

in Caodeng rGyalrong is not merely a structurally motivated case, since the marker
can be eliminated without this resulting in ungrammaticality. The primary function
of the ergative marking of Agent is to emphasize the somewhat unusual semantic
role assignment in (80). Differently from Finnish, the accusative is not a mere
structurally motivated case in Turkish, since it can be omitted, as in (82). The
primary function of the accusative is to highlight the definiteness of the Patient.
The structural vs. semantic marking (in the sense the labels are used here)
illustrated above differ from each other also in the distribution of the features. The
sole explicit marker of transitivity is responsible for the expression of transitivity in
general in structurally motivated cases (cf. 4.5.2.3. below). It is very difficult (or
even impossible) to distinguish between different features. For example, in Finnish
the accusative marking of the Patient is the sole indicator of transitivity. It is not
only related to the affectedness of the patient, but the use of the construction implies
that the overall transitivity of the event profiled is high. On the other hand, in
Caodeng rGyalrong the ergative marker is only optional and it might be omitted.
The event denoted in (79) is transitive, since it involves two participants, one of
which instigates the event, whereas the other is affected by it. The ergative marker
is not related to the expression of transitivity in general, since it can be omitted
without this resulting in ungrammaticality or intransitivity. In (79) and (80), the
basic indicator of transitivity is the reference to two participants. Ergative marking
of the Agent can be seen as ‘extra’.
In cases illustrated in (76)-(82), the distinction between structural and
semantic marking is rather indisputable. However, the overall picture is not this
clear. This is discussed in light of the following examples from Finnish (cf. also
(77)-(78)), cf.

(83) mies jo-i maido-n

man.NOM drink.PAST-3SG milk-ACC
‘The man drank the milk (up)’
(84) mies jo-i maito-a
man.NOM drink.PAST-3SG milk-PART
‘The man drank (some) milk’
(85) mies viha-si maito-a (*maido-n)
man.NOM hate-3SG.PAST milk-PART (milk-ACC)
96

‘The man hated milk’

Example (83) seems at first sight similar to (77). However, obvious differences
arise, if we compare (83) to (84). Since the Patient of juoda (as well as that of many
other verbs) can also appear in the partitive, the accusative marking in (83) is not
solely structural, but should rather be labelled as semantic. In (83), the accusative
marking has the function of marking the perfective aspect. What makes the picture
even messier is that the partitive marking can also be semantically motivated as in
(85). The marking is not related to the mere lower degree of affectedness, but the
event is less transitive in (85). Examples like (83)-(85) show that the division into
structural vs. semantic motivation does not necessarily classify languages, but rather
different transitive clauses.
In addition to differences between semantically and structurally conditioned
marking types, it is important to note that the marking may be related to either a
high or a low degree of transitivity (as noted above). Examples (83)-(85) from
Finnish illustrate both types. We do not want to restrict the use of the term
‘transitivity marking’ only to clauses displaying high degrees of transitivity, since
also marking of low (or decreased) transitivity is motivated by transitivity. In the
case of explicit marking of lower degrees of transitivity, deviations from the basic
scheme are viewed as sufficiently relevant to be marked explicitly. This kind of
marking is also motivated by transitivity, bu the determining factor is different.
Marking of high or low transitivity can be related primarily to the verb or to the
arguments. Both of these are discussed below. We start by illustrating the marking
of transitivity on the verb. In these cases the marking is usually valency-conditioned.
The illustration is based on Haspelmath’s 1993(b) paper. In addition to changes in
valency, the marking may also be motivated by inherent semantics of events. The
argument marking is more important in this case.
Haspelmath (1993b:91) has distinguished four different types according to
whether low or high transitivity, neither or both are signalled on the verb. The
division is based on 31 verb pairs from 21 languages. The conditioning factor is how
certain transitivity alternations are marked. If a transitive verb (e.g. ‘kill’) is derived
from an inchoative one (e.g. ‘die’), we are dealing with a causative alternation. The
 97

opposite cases illustrate anticausative alternations. Third, there are non-directed

alternations, like English ‘cook (TRANS)’ vs. ‘cook (INTR)’. There are no
morphological differences , even if the transitive valence has clearly been affected
(consequently, the only difference is the number and marking of arguments). In
equipollent alternations, both transitive and intransitive verbs are marked, e.g.
Japanese atum-aru (‘gather’ (INTR)) vs. atum-eru (‘gather’ (TRANS)). Already this
very brief illustration suffices to show that there is no unilateral direction of
derivation and depending on the language and individual verbs, either high or low
transitivity can be explicitly marked. It would not make justice to the cross-linguistic
variation to state that we are dealing with marking of transitivity only if the high
transitivity is explicitly marked. Both of these marking types are equally important.
The classification of Haspelmath is based on the relation between verb
morphology and valence of verbs. The typology proposed by Haspelmath can also
be applied to argument marking, which is illustrated below, cf.

(86) mawun-du gu ar bu õ-an

man-ERG dog hit-PRES
‘A man hits a dog’
(87) mawun `a  u ma -an
man Djaru talk-PRES
‘A man talks Djaru’ (Djaru, Tsunoda 1987: 97)

(88) lam cya wælikandak hæduwa

child.NOM sand.hill.INDEF make.PAST
‘The child makes a sandpile’
(89) lam cya-T c eek c æhuna
child-DAT that hear.PAST
‘The child heard that’ (Sinhala, Gair 1998b: 68, 1998: 30)

(90) §ò § p c§ ká §
I eat fish
‘I eat fish’
(91) §ò § õ]§ s ]§
I afraid dog
‘I am afraid of the dog’ (Kammu, Svantenson 1983: 77)
98

(92) e t~ le toe e lona tinana

ASP hit ART child PREP (‘ERG’) her/SG mother
‘The mother beats her child’
(93) e alofa le tinana ki ona toe
ASP love ART mother PREP (‘ACC’) her.PL child
‘The mother loves her children’ (Futunian, Lazard 1998:46, cited from Moyse-Faurie 1991:
71f)

Above, the marking of arguments refers to marking patterns, not to the marking of
individual arguments. Examples (86) and (87) illustrate the equivalent of
Haspelmath’s causative type. In this case the more transitive clause is the more
explicitly marked one. Examples (88) and (89) from Sinhala, on the other hand,
exemplify explicit marking of less transitive events, i.e. the ‘anticausative’ type.
Kammu illustrates a non-directed type, since the marking of both highly transitive
and less transitive clauses is the same. In Futunian, both highly transitive and less
transitive events are explicitly marked, but by different marking patterns.

4.2. Linguistic vs. semantic marking

The marking of transitivity may be either linguistic (explicit) or semantic (implicit)

(the latter must not be confused with semantically motivated transitive marking
discussed in 4.1.). Both labels refer to linguistic manifestation of non-linguistic
transitivity, but their nature is different. The former label refers to cases in which
non-linguistic transitivity is marked explicitly with linguistic mechanisms, i.e.
transitivity of non-linguistic events is marked morphosyntactically. Semantic
marking comprises cases in which the transitivity is not marked explicitly by
linguistic mechanisms, but our interpretation of events as transitive is primarily
based on the semantics of the verbs. Both these types of transitivity marking are
illustrated in ligh of concrete examples below.
Linguistic marking of transitivity can be understood in two different ways.
First, it refers to cases in which basic transitivity of events is marked explicitly. This
kind of marking is often redundant, since the right interpretation can be inferred
from the semantics of verbs without any explicit marking. This kind of linguistic
 99

marking will be later referred to as explicit marking, since aspects we are familiar
with are marked explicitly. Examples include the following, cf

(94) mies rikko-i ikkuna-n

man.NOM break-PAST.3SG window-ACC
‘The man broke the window’
(95) mies viha-a musiikki-a
man.NOM hate-3SG.PRES music-PART
‘The man hates music’ (Finnish)

(96) cuo vz vu:

he.ERG he.NOM kill
‘He kills him’
(97) suo:na yz vie:z
I-DAT he.NOM like
‘I like him’ (Ingush, Nichols 1994b: 119f)

(98) õangula-yu algi- õ a `i wagu:ray

he-ERG kill-PAST this.ABS kangaroo.ABS
‘He killed this kangaroo’
(99) a `i bu `u:r wayi-`i daba:y-a
this boy.ABS afraid-PRES dog-LOC
‘The boy is afraid of the dog’ (Yaygir, Crowley 1979: 374)

The former example of each pair designates a highly transitive event, whereas the
transitivity of the latter is clearly lower. The argument marking varies accordingly.
Explicit marking would not be necessary in these cases, since we can infer the
semantic roles of the participants from the inherent semantics of verbs. On the basis
of our knowledge about the events in the non-linguistic world, we know that the
events denoted in the former examples are transitive, whereas those in the latter
examples are not.29 As can be inferred from the examples given above, explicit (as
well as linguistic) marking of transitivity is highly language-specific. Languages
differ crucially from each other in what kinds of change in the inherent semantics

29
Even if we are not aware of the concept of ontological transitivity, we know that these two types
of event are distinct.
100

of events result in a linguistic change.

As noted, the explicit marking in cases like those illustrated above usually
does not reveal anything about the nature of events that would not inhere in the
semantics of verbs. In a way, explicit marking means that the structure adjusts itself
to the semantics of events profiled. Or, put another way, the transitivity of the verb
‘radiates’ to its arguments. In these cases, the use of the basic transitive clause is
semantically motivated and it may be used only if the semantic transitivity of events
is sufficient. We associate high ontological transitivity with this pattern.. If the
semantic transitivity of a verb is considerably lower than in typical transitive cases,
the structure has to be changed in order to avoid mismatches of structural and
semantic transitivity. In (94)-(99), there is no disagreement, since the structure is
manipulated according to the semantic transitivity of events.
The other facet of linguistic marking covers cases in which semantically non-
inferable changes in transitivity are marked. This differs from the previous type in
that the marking is by no means semantically redundant. The highlighted differences
cannot be inferred from the semantics of verbs, but the explicit marking is needed
in order to secure the intended meaning. The marking usually emphasizes
deviations from the basic scheme. This kind of marking directly affects the
interpretation of clauses and their transitivity. Examples are illustrated below, cf.

(100) ngatyu-ngku nga-rna-Ø ngumpirr-Ø pat-man-i

I-ERG AUX-1SGI-3SGII woman.ABS touch-do-PAST
‘I touched the woman’ (action completed)

(101) ngatyu-ngku nga-rna-la ngumpirr-la pat-man-i

I-ERG AUX-1SGI-3SGIII woman.DAT touch-do-PAST
‘I tried to touch the woman’ (action attempted) (Djaru, Tsunoda 1994: 4676)

(102) ngarritj-thu nhä-ma wäyin

subsection term-ERG see-UNM animal.ABS
‘Ngarritj sees the animal(s)’
(103) ngarritj nhä-ma wäyin-gu
subsection term.ABS see-UNM animal-DAT
‘Ngarritj is looking for animal(s)’ (Djapu, Morphy 1983: 38)
 101

(104) ti he-v
tree chop-PRES
‘Chop down a tree’
(105) ti-m he-the-v
tree-DAT chop-INTR-PRES
‘Chop on a tree’ (W aris, Foley 1986: 109)

Examples (100)-(105) illustrate in principle the same events, but with certain
significant differences. These aspects are not readily inferable from the basic
semantics of the verb, even if the semantics of the event allows the variation
illustrated. In the typical cases (i.e. in the former example of each pair), clauses are
interpreted according to basic semantics of events. If the intended interpretation
differs from this, transitivity must be manipulated for the purpose of avoiding
misinterpretations. Without the non-basic marking, we interpret the clause on the
basis of its inherent basic semantics, i.e. the marking illustrated in (100)-(105)
follows from deviations from the basic transitivity of events or verbs. This kind of
marking is here referred to as linguistic in order to distinguish it from the explicit
marking illustrated in (94)-(99).
In (100)-(105), either the basic transitivity of events or significant changes in
it are marked morphosyntactically. However, as stated above, the explicit marking
is in many cases redundant as regards the marking (or interpretation) of transitivity
per se. The marking in (94)-(99) does not aid us in interpreting the clauses correctly,
since the interpretation is primarily based on the semantics of verbs and on our
knowledge about the nature of events. There are also languages in which an explicit
marking is much more restricted and does not usually appear. This kind of
expression of transitivity is here understood as semantic (or implicit), since it is the
opposite of explicit marking. The marking occurs only on the verb without any
morphosyntactic mechanisms. Typical examples of languages in which the marking
of transitivity is merely implicit include analytic languages, the structure of which
does not enable explicit marking of transitivity with e.g. morphological cases.
Hence, the implicitness is usually not an option, but it is a result of the structure of
languages. Examples of what is here regarded as implicit marking are given below,
cf.
102

(106) §ò § p c§ ká §
I eat fish
‘I eat fish’
(107) §ò § õ]§ s ]§
I afraid dog
‘I am afraid of the dog’ (Kammu, Svantenson 1983: 77)

(108) man-nen döda-de hund-en

man-DEF kill-PAST dog-DEF
‘The man killed the dog’
(109) man-nen älska-de hund-en
man-DEF love-PAST dog-DEF
‘The man loved the dog’ (Swedish)

Kammu and Swedish both lack morphological means to mark case on nouns. This
results in what is here labelled as implicit marking. Even if the differences in
transitivity are obvious, the marking remains constant. This does not, however,
result in any misinterpretations, since we are dealing with the typical transitivity
inhering in the semantics of verbs. Since the right interpretation of events is possible
without any explicit marking, we may conclude that the kind of marking attested in
these examples is more economical than those exemplified in (94)-(99). However,
in languages like Kammu and Swedish the implicit marking is not an ‘option’, since
the highly analytic structure of these languages makes it impossible to mark
semantic differences illustrated above explicitly.

4.3. Direct vs. indirect marking

Above, we have distinguished different kinds of transitivity marking based, for

example, on the nature of transitivity marking (is it semantic or linguistic). In this
section, we illustrate and discuss what we have labelled as direct vs. indirect
marking. The distinction between these two types is based both on structural and
semantic criteria. First, the types differ from each other in how salient the marking
is. In the case of direct marking, the marking is very directly observable and it
appears on linguistic elements whose primary function is to express transitivity
 103

(these include case markers and different verbal particles). In the case of indirect
marking, on the other hand, the marking appears on elements that are usually not
considered relevant for transitivity (typical elements include adverbials and TAM-
markers). Second, direct and indirect marking can be distinguished due to their
contribution to the interpretation of transitivity. Our interpretation of transitivity is
primarily based on direct marking. This means that we regard constructions as more
or less transitive based on elements of direct marking. The indirect marking is less
important in this respect. Mechanisms of indirect transitivity marking make only a
minor contribution to the expression of transitivity, since they are primarily related
to the expression of other functions. This means that they are never alone
responsible for whether clauses are interpreted as transitive or intransitive. They are
rather regarded as elements that are obligatorily adjusted to transitivity of clauses.
In this section, we will focus on the illustration of indirect marking. There are
several reasons for doing this. First, indirect marking is clearly the less studied of
the types. Second, direct marking has already been (even implicitly) illustrated so
far. And third, a more detailed study of indirect marking may provide such
perspectives that would not be discovered, if we focussed on the illustration of
direct marking. The direct marking is discussed and illustrated only in light of few
typical examples. We begin by discussing the more familiar direct marking briefly.
All the examples illustrated so far have exemplified direct transitivity
marking. This includes all the marking on the verb or arguments that directly
signals some aspect of transitivity. Since we are dealing with a direct marking, this
is very natural, because these elements refer to the event itself (verb) and the
participants involved (arguments). Other elements of clauses are usually only
indirectly (if at all) related to transitivity. Hence, the directness is due to the
appearance of the relevant markers on the constituents that signal transitivity most
directly, cf.

(110) nainen sö-i puuro-n

woman.NOM eat.PAST-3SG porridge-ACC
‘The woman ate the porridge’ (Finnish)
104

(111) hansi-p inuit tuqup-paa

Hansi-ERG people kill-IND.3SG.3SG
‘Hansi killed the people’ (W est Greenlandic, Manning 1996: 82)

(112) ngaju juga-ala nyabay

I.ERG drink-PRES water.ABS
‘I am drinking water’
(113) ngay juga-le-la nyabay
I.NOM drink-ANTIP-PRES water.ABS
‘I am drinking water (repeatedly)’ (Bandjalang, Austin 1982: 39)

Examples (110)-(113) exemplify very typical instances of direct marking. The

examples differ from each other in what aspect of high transitivity the marking is
most directly related to. In (110) from Finnish, the explicitly marked facet is the
(total) affectedness of the patient and the completion of the event. In (111), the high
degree of agency is emphasized by marking the Agent in the ergative. Furthermore,
the overall high transitivity of the event is marked on the verb through the form of
the cross-reference affix. Examples from Bandjalang exemplify direct marking of
both high and decreased transitivity. In (112), the most directly observable
mechanism of transitivity marking is the ergative marking of Agent. In (113), the
decreased transitivity is signalled by manipulating the marking of the Agent and the
morphology of the verb.
We can distinguish two different types of transitivity marking both of which
can with reason be labelled as indirect. First, the indirectness of the marking may
be due to the primary function of the marking. In this case, some element serves
primarily some other function, in addition to which it is also related to transitivity.
Typical examples of this include tense and mood markers on the verb. Secondly, the
indirectness may follow from the ‘morphosyntactic location’ of the marking. As
exemplified above, direct marking is always somehow related to either arguments
or the verb. The indirect marking, on the other hand, can appear, for example, on
adverbials or other (non-core) constituents. In the latter cases, the direct and indirect
marking can be distinguished formally. Both of these types are illustrated and
discussed in what follows.
In (112) and (113), direct transitivity markers attach to the verb. We know that
 105

in (113) we are dealing with a less transitive clause, since an antipassive affix is
attached to the verb. However, not all facets of verb are equally important as regards
the marking of (semantic) transitivity. Many categories that are only indirectly (if
at all) related to transitivity are also expressed typically on the verb. Tense and
mood are examples of this.30 The indirectness refers here to the form of these
markers that may vary according to the transitivity of clauses. If these markers are
used to signal transitivity, the marking may not be considered as direct as it is in
genuinely direct cases, cf.

(114) àng wà-apm-ì

3SG say-TMdys-3.INTR.PAST
‘He said something’
(115) à:ng-í wà-apm-à
3SG-AGT do-TMdys-3.TR.PAST
‘He did something’ (Dulong/Rawang, LaPolla 2000: 293)

(116) õad’a wagun ganda-ñu

I.ERG wood.ABS burn-PERF.TR
‘I burnt some wood’
(117) õayba ganda-gi
I.NOM burn-PERF.INTR
‘I burnt myself’ (W argamay, Dixon 1981: 64)

As can be seen, the form of the tense marker varies according to the (semantic)
transitivity of the designated event. There is both direct and indirect marking
involved here. In both languages, transitivity is directly manifested in the ergative
marking of the Agent. Moreover, transitivity is (indirectly) reflected in the selection
of the tense marker that is sensitive to transitivity. The latter can rightfully be
regarded as indirect functionally, because an element that is only indirectly related
to the most significant aspects of transitivity reflects the transitivity of events.
Deviations are deemed ungrammatical. In order that the tense affix could be labelled
as a direct marker, there should not be any direct marking involved, i.e. we should

30
It is order to emphasize here that we do not wish to say that tense and mood would be irrelevant as
regards the expression of transitivity (see below for examples).
106

be able to express the kinds of alternation illustrated here solely by manipulating the
form of the tense affix.
A very similar example of indirect marking is illustrated below, cf.

(118) thadna-ru
leave-IM P
‘Leave it alone’
(119) kathi-ra
turn.back-IM P
‘Turn back’ (W angkankurru, Hercus 1994: 180f)

In this case, the relevant element is the imperative suffix. (118) and (119) differ
from the examples from Dulong/Rawang and Wargamay in that there is no direct
marking involved, but the differences in transitivity (transitive valence) are only
implicit. However, in this case the different form of the imperative suffix is not
primarily responsible for our interpretation of transitivity, either, but we base our
interpretation primarily on semantics.
Examples (114)-(119) illustrate cases in which certain elements on verbs are
regarded as indirect markers, even if, as noted above, direct marking is typically
related to the verb. The indirectness is here based on the fact that the affixes in
question are not primarily related to the expression of transitivity and that our
interpretation of transitivity is not based primarily on them. Furthermore, these
markers do not affect the valence of the verb, but they must be adjusted to it. For
example, the antipassive affix in (113) is responsible for the decreased transitivity
of the exemplified clause, while in (117) the tense affix has been affected by the
decreased valence.
In addition to (114)-(119), in which the differences between direct and indirect
marking are not necessarily obvious, there are clear cases. Typical examples include
transitivity marking on adverbs of different kinds, cf.

(120) pama-ngku warrngu-Ø mayka-n yama-nga-n

man-ERG woman.ABS tell-P/P so-TR-P/P
‘The man told the woman so’
 107

(121) pama-Ø yama-pi-n mayka-kali-n

man-ABS so-INTR-P/P tell-AP-P/P
‘The man said so’ (W arrungu, Tsunoda 1988: 599)

(122) jankarra-n yaparru-lu ngali puni-kurrara

tie-IMPER quick-ERG we.DL.NOM go-INTENT
‘Tie it up quickly so we can go’ (Thalanyji)
(123) pampa-ngku-nthi ngatha mama-lkurni-a
cannot-ERG-just I.ERG get-CONT-PRES
‘I just can’t get them’ (Jiwarli, Austin 1995: 16) 31

(124) wao §i§-kiran-i haw-i

cry-coming(INTR)-CONJ come-CONJ
‘He came crying’
(125) his-b vnan-kin bi m v§v-yama-kin
see-coming(TR)-CONJ but touch-NEG-CONJ
‘He saw it as he came but didn’t touch it’
(126) hano §-a Õ haw-i
there-LOC(INTR) come-CONJ
‘He came from there’
(127) hano-Õon his-kin
there-LOC(TR) see-CONJ
‘He saw it from there’ (Capanawa, Loos 1999: 239)

(128) hatik-a Õ-bi na •i-kan-a iki

altogether-INTR-EMP bathe-PL-CMPL AUX
‘The bathed altogether’
(129) hatí-Õon-bi na •i-ma-kan-a iki
altogether-TRANS-EMP bathe-CAUS-PL-CMPL AUX
‘The bathed (him/her) altogether’ (Shipibo-Conibo, Valenzuela 2001: 1)

(130) a õki-§a dzu-kh c§-k-a

we(EXCL)-ERG eat-take.away(ITIVE)-1PL-ERG
‘W e ate it up’

31
Examples (122) and (123) are taken from reading pack for Peter Austin’s class ‘Aboriginal and
Torres Strait Islander Languages of Australia’ held at LSA 2001 Institute at Santa Barbara. The page
numbers do not match with the original ones, since I do not have access to the original reference.
108

(131) hil-khuts-a
mix-go(ITIVE)-23SG
‘It’s totally mixed up’ (Dumi, Van Driem 1993: 210f)

Above, the indirectness of marking is more obvious than in (114)-(117). Transitivity

is signalled indirectly on completely optional adverbs that are marked according to
the transitivity of clauses. Hence, in case the adverbs are omitted transitivity is
marked only directly. Examples above illustrate so-called transitivity concord. The
term is here adopted from Loos (1999:239) and it is also used for examples from
Australian languages and Dumi. In the case of (adverbial) transitivity concord, an
element that is usually neutral for transitivity, contributes to the overall transitivity
of clauses. For example, in Warrungu the form of the affixes on adverbials is
sensitive to transitivity. The case is very similar in Thalanyji and Jiwarli, two other
Pama-Nyungan languages in which the ergative affix attaches to manner adverbs or
root modals only in transitive clauses. It cannot appear in intransitive clauses.
Formally, the ergative affix that appears on manner adverbials or root modals is
identical to the typical ergative affix attached to nouns. Only the location is
‘untypical’. The case is somewhat different in Capanawa, since adverbials appear
in different forms conditioned by the transitivity of clauses. The same applies to
locatives, as can be seen in (126) and (127). This kind of marking clearly is
functionally indirect, since whether someone comes from place rather than place B
is irrelevant for transitivity. Shipibo-Conibo (as illustrated by Valenzuela) is another
language with an extensive transitivity concord (other examples include Yaminawa,
Chakobo and Pakawara, as illustrated in Valenzuela 2001). Examples from Dumi
exemplify instances in which the itive aspect marker agrees with the transitivity of
the clause. These cases differ somewhat from others under study here. As generally
known, the aspect makes a very significant contribution to transitivity. Clauses in
perfective aspect usually receive a more transitive marking (if there is a difference).
However, we may rightfully claim that (130) and (131) illustrate indirect marking
of transitivity. It is indisputable that the marker khc§ in (130) emphasizes the total
affectedness of the patient (and the perfective aspect) of the event. In (131), khc§ is
a direct marker of direct transitivity. The indirectness of the marking becomes
obvious in comparison with (131). Both (130) and (131) are in the perfective aspect.
 109

If the itive aspect marker were a typical marker of direct transitivity in Dumi, we
would not expect to find the differences exemplified above. The distinction between
explicit perfective aspect vs. zero marking reflects transitivity directly and suffices
to underline the relevant transitivity feature. Since, however, the form of the marker
is sensitive to transitivity of clauses, we are dealing with indirect marking here.
Previously we have been illustrating examples of indirect marking in which
the indirectness has been quite obvious. In (114)-(119), transitivity is marked
indirectly on the verb through the form of tense or imperative suffixes. In these
cases, the indirect markers are parts of the finite verb that in many cases is primarily
responsible for the direct marking of transitivity. In (120)-(131), on the other hand,
the location of the marking is different from that of direct marking. Moreover, there
are less clear cases in which it may be difficult to say whether we are dealing with
a direct or an indirect marking. A few problematic examples of this are illustrated
(and discussed) below, cf.

(132) ngi-rim-fifi
1SGS-Sit-smoke
‘I’m smoking’
(133) ngi-nyinggin-nyi-kerrety
1SGA-See-2SGO-watch
‘I’m looking after you’
(134) nge-rim-ø-pal
1SGA-Hands-3SGO-break
‘I broke it’
(135) ye-nim-pal
3SGS-Go-break
‘It is broken’ (Ngan’gityemerri, Reid 2000: 340, 347)

(136) t’crmus-u s vbb vr alc

bottle-DEF break.VN say.PERF.3M
‘The bottle broke’
(137) *aster t’crmus-u-n s vbb vr alc-
A. bottle-DEF-ACC break.VN say.PERF-3F
(for: Aster broke the bottle)
(138) aster t’crmus-u-n s vbb vr ad crr cg c-
A. bottle-DEF-ACC break.VN do.PERF-3F
110

‘Aster broke the bottle’ (Amharic, Amberber 2000: 328f)

(132)-(138) illustrate cases in which the overall transitivity of clauses follows from
the ‘co-operation’ of two different kinds of verb that can be classified as finite verbs
and coverbs.32 We may also say that the lexical verbs in (132)-(138) involve both
a preverb and an inflecting verb that together mark transitivity. In examples cited
above the (infinite) coverbs are responsible for the basic nature of events. However,
in Ngan’gityemerri and Amharic, the semantically meaningful infinite verbs cannot
alone mark the overall transitivity of events, but finite verbs must co-occur with
these elements. For example, in Ngan’gityemerri, person affixes can be attached
only to finite verbs and not to lexical coverbs. The coverbs must also take account
of the transitivity of the described event. In both cited languages the coverb and the
finite verb must agree in transitivity. For example, in (132), the finite verb is
intransitive (‘sit’), while in (133), it is transitive (‘see’). This must be the case, since
‘see’ allows two participants referred to and this is required by the semantics of the
event in (133). Deviations are deemed ungrammatical. In this case, we are clearly
dealing with a kind of transitivity concord, since the finite verb simply has to be
adjusted to the transitivity of the lexical coverb and it does not make any significant
contribution to the transitivity of events. If we regard direct marking to be associated
with the elements most responsible for the description of events and the
interpretation of transitivity, we cannot but label examples like (132) and (133) as
instances of indirect marking. All the relevant marking appears on the finite verb.
However, as can be seen in (134)-(138), finite verbs are in some cases the sole
markers of valencial changes. In these cases, it is justified to say that we are dealing
with a direct marking, since the marking appears on the element that is most
responsible for the marking of valencial changes. Consequently, according to the
verb and construction in question our analysis varies. If we wish to do justice to the
data, we can perhaps label the kind of marking as distributed, since both verbal
elements contribute significantly to the transitivity marking, and it is very difficult
to argue justly for adequacy of either analysis.

32
The labels are adopted from Reid (ibidem).
 111

4.4. Primary vs. secondary marking of transitivity

The relation of markers employed in the marking of transitivity and events denoted
may be either direct or indirect (this division has to be distinguished from the
direct/indirect marking division illustrated above). In the former case, the primary
function of the marker is to mark the (either high or low) transitivity of events by
either emphasizing some (individual) aspect associated with transitivity or by
pointing out to the high overall transitivity of the event in question.33 In the case of
secondary marking, the marking is also related to transitivity, but it is primarily used
in the expression of other functions that may be, for example, the disambiguation
of the semantic role assignment. The employed mechanisms may be (and usually
are) formally identical, i.e. the markers are distinguishable only on the basis of the
underlying motivation. The primary function may also be to signal structural
transitivity. This is the case only if an obvious semantic motivation is lacking. The
primacy of structural motivation is merely a default choice in the absence of any
semantic motivation. All of the briefly introduced functions of transitivity marking
are illustrated and discussed in what follows. We start by illustrating the primary
marking. After this we proceed to cases in which the marking is not motivated
primarily by transitivity but by other function(s).
In order that we are entitled to speak of a primary transitivity marker, we have
to be able to show that the marking is uncontroversially related to the expression of
an individual, significant aspect of transitivity or to the overall transitivity of events.
If the marker is omitted, the given feature is not understood as a part of the meaning
of a clause. Nor can it appear, if the transitivity is significantly lower (in case high
transitivity is explicitly or linguistically marked). Usually, this kind of marking is
primarily semantically determined, and syntax only makes a minor contribution.
Typical examples of primary marking involve the following, cf.

(139) ada jak t’üna

he.ERG meat.ABS eat.AOR
‘He has eaten meat’

33
The differences between these are not always clear.
112

(140) ada t’üna

he.ERG eat.AOR
‘He has eaten’ (Lezgian, examples courtesy of Martin Haspelmath)
(141) zun ata-na
I.ABS come-AOR
‘I came’
(142) za-z sa ktab ž —a-na
I-DAT one book find-AOR
‘I found a book’ (Lezgian, Haspelmath 1993: 251, 281)

(143) darj § m ~riyo giyo

tailor killed went
‘The tailor was killed’
(144) darj § ne m ~riyo giyo
tailor ne killed went
‘The tailor was murdered’ (Marwari, Magier 1990: 217)

Examples (139)-(144) illustrate two different kinds of primary marking. In (139),

the ergative marking of the Agent follows from the high overall transitivity of the
event profiled. As can be seen in (140), the marking is not motivated by structural
transitivity, since the marking appears also in structurally intransitive clauses as long
as the number of participants involved is two. The primacy is further underlined by
the fact that the marking is restricted to semantically transitive events. The marking
changes, if the event profiled is intransitive or, if the degree of transitivity of the
event is very low, as is shown in (141) and (142). In Lezgian, the omission of the
ergatively marked argument does not necessarily result in the omission of the
transitivity altogether. Examples from Marwari are different in this respect. The
primary function of the particle ne is to signal high agency and transitivity and the
omission of the marker coincides with the omission of the semantic feature
denoted.34 The number of arguments is irrelevant in this case, since ne can appear,
even if the argument that refers to the relevant participant, would have been omitted.
The other type of primary motivation is illustrated by cases in which the
transitive marking is conditioned purely or primarily by syntactic transitivity. In

34
As pointed out by Magier (1990:218) this particle is best viewed as a general transitivizing particle.
 113

these cases, the marking is not conditioned by high ontological transitivity, but the
mere existence of two arguments suffices, cf.

(145) k vp vk te m «m ka §ok mep

axe ERG wood hard cut
‘The axe cuts hard wood’ (Maxakalí, Rodrigues 1999b: 194)

(146) jaakko kuul-i ihmeääne-n

Jacob.NOM hear-PAST.3SG strange.sound-ACC
‘Jaakko heard a/the strange sound’ (Finnish)

Events above are both denoted by the basic transitive pattern of the given languages
despite the very low degree of semantic transitivity involved. Consequently, we
cannot argue justly for a semantic motivation of the marking. The only possible
motivation here is the transitive valence.
The most typical examples of marking that can be considered secondary are
perhaps provided by instances of differential object marking (Lazard 1998: 82 and
elsewhere). As generally known, in many languages definite patients are marked
differently from indefinite ones. Since this is not directly related to semantic (or
rather ontological) transitivity, the marking is best regarded as secondary (cf. also
(79)-(82) above). A couple of further examples of secondary marking are illustrated
below, cf.

(147) nuu pei o’o-maaka

3SG canoe make-NEAR.PAST
‘He made a canoe’
(148) nu-ro tiramu ea’a-maaka
3SG-SUBJ Tiramu see-MID.PAST
‘He saw Tiramu’ (Kope, Clifton, 1995: 54f)

(149) ap wekki wat-n-an-h-e

man charcoal hit-1SG-put-REAL-3SG.A
‘The man smeared charcoal on me’
(150) ap palu-nen Ø-na-sikh-e
man python-ERG 3SG.U-eat-RM.PAST-3SG.A
‘The python ate the man’ (Dani, Foley 1986: 107)
114

Also in (147)-(150), the primary function of the marking seems to be to signal

transitivity. However, differently from (139)-(142), for example, the appearance of
the ergative marker is not conditioned by semantic transitivity, but the primary
function of the ergative marking of Agent is to disambiguate clauses. In Kope and
Dani, the ergative marking of Agent is obligatory only if both participants involved
are possible agents, i.e. if non-linguistic evidence does not disambiguate the role
inventory. Consequently, the Agent is marked ergatively in (148) and (150), but not
in (147) and (149). Since the ergative marking can be eliminated, even if the event
profiled is highly transitive, the marking cannot be motivated by structural nor
semantic transitivity. Furthermore, in languages like Dani the ergative marking is
restricted to less typical agents, i.e. to less transitive cases (see also the discussion
in Comrie 1978: 355ff). The kind of marking exemplified above illustrates the
discriminatory function of argument marking (cf. e.g. Comrie 1975, 1978, Dixon
1979, Song 2001: 156ff)35, since the marking appears only in case there is a need to
distinguish between participants linguistically.

4.5. On the basic transitive clause

4.5.1. Defining the basic transitive clause

In the previous section, we briefly illustrated and discussed what is here labelled as
primary vs. secondary transitivity marking. The goal was to show that high transitive
marking is not necessarily always based on any facet of semantic or structural
transitivity, but other aspects contribute to the marking as well. Consequently,
transitive marking does not necessarily signal high transitivity. In this section, we
proceed to illustrating the semantic and structural nature of basic transitive clauses.
We start by illustrating the concept of basic transitive clause briefly, after which we
proceed to discussing the underlying motivation of transitive marking.
In order to propose a typology of the basic transitive construction, we have to
start by defining what we mean by the notion under study. In a nutshell, the notion

35
In the present context the term is used somewhat differently.
‘basic transitive clause’ refers here to the structure that is employed in the
description of semantically highly transitive events (as defined above) in a given
language. The event involves an agent and a patient and both participants are
somehow explicitly referred to (not necessarily by independent (nominal)
arguments). Transitive clauses can always be distinguished from (valencially)
intransitive ones based on this: intransitive clauses profile events that involve one
participant only. The concept ‘basic transitive clause’ is highly language-specific.
First, languages differ enormously from each other in the form of the structure in
question. The only truly universal structural feature of basic transitive clauses is the
explicit reference to two participants (which is also optional in many languages).
Furthermore, languages diverge in what kinds of event can be denoted by the basic
transitive clause. Put concretely, this means that in language A only highly transitive
events are expressed by this construction, whereas in language B the variety is
greater (see, e.g. Tsunoda 1985:387 and Drossard 1991:411ff). However, as long as
the structure remains the same we are entitled to label the structure in question as
a basic transitive clause (cf. Kozinsky and Lazard). There are, however, some
restrictions. In many (most?) languages the participants of transitive events are
somehow explicitly separated from each other. For example, in predominantly
accusative languages the Patient appears in the accusative case and the nominatively
marked argument is by default understood to refer to the agent, due to the semantics
of a given event. In a way, both arguments are marked according to their semantic
roles in transitive clauses (one explicitly, the other by default). This relation must
always hold in transitive clauses, even if the degrees of agency and/or affectedness
were reduced, i.e the agentively (explicitly or not) marked argument must refer to
some kind of instigator and the Patient must refer to the less active participant in a
given event (the marking of the roles has to be based on the marking of Agent and
Patient in typical transitive clauses). If this is not the case, we cannot speak of
typical transitive clauses. A possibly problematic case is illustrated below, cf.

(151) mú "tu §a " hk’úm

3.PAT 1.AGT kill
‘I killed him’
(152) to " qadálmada mu "l
1.PAT hate 3.AGT
‘He hates me’
116

(153) mu "l qadála to " §ú "daw

3.AGT hate 1.PAT really
‘I really hate him’ (Central Pomo, examples courtesy of Marianne Mithun) 36

Of these examples, only (151) and (152) are here regarded instances of the typical
transitive clause of Central Pomo. Example (151) is transitive semantically and
(152) is structurally similar to it despite the lower degree of transitivity. In these
cases, the Agent is an agent, whereas the Patient refers to the participant affected by
the event denoted, i.e. the marking correlates with the semantic role assignment. As
regards the employed mechanisms of argument marking, (153) is identical to both
(151) and (152). However, it is here excluded as a typical transitive clause of
Central Pomo because of its deviant semantic role assignment (which is also
signalled on the verb). In (153), the argument that usually refers to the instigator of
the event, refers to the participant that is rather considered as the target and vice
versa. The energy flow (see Langacker 1991: 327f) is therefore different from that
implied by the argument marking. Consequently, the semantic relation noted above
does not hold.
Even if the proposed definition as such seems applicable universally, it is far
from being without problems. First, it is very difficult to give an exhaustive
semantic (non-structural) definition of a basic transitive event, since the concept
involves so many different aspects (even if the kind of definition proposed in 3.2.
suffices in most cases). Second, not only semantics, but other factors determine the
use of the structure as well. This means that even if we were able to define the basic
transitive event exhaustively, we still cannot be sure, whether this structure
exemplifies the basic transitive structure in a given language or not. The two
participants in a transitive event can in many languages be referred to in more than
one way, and it may in some cases be difficult to determine which of these we can
justly regard as basic. If the different marking is based on differences in semantic
transitivity or it can be explained by referring to semantics indirectly, this is not a
problem. For example, in the following example it is safe to say that (154) and

36
All examples marked as courtesy of Marianne Mithun are from handouts given by Professor Mithun
in her class ‘Grammatical relations, argument structure, case and voice in North American Indian
languages’ at Linguistic Institute 2001 at UCSB.
 117

(156), and not (155) or (157) should be regarded as typical transitive clauses, cf.

(154) er töte-te den mann

he.NOM kill-3SG.PAST ART.ACC man
‘He killed the man’
(155) er dachte an den mann
he.NOM think.PAST.3SG PREP ART.ACC man
‘He thought about the man’ (German)

(156) jaakko sö-i kalapuiko-n

jacob.NOM eat-PAST.3SG fish stick-ACC
‘Jaakko ate the fish stick’
(157) jaakko katso-i teletappe-ja
Jacob.NOM look.at-PAST.3SG Teletubbies-PART
‘Jaakko was watching Teletubbies’ (Finnish)

If we assume that high semantic transitivity correlates with high structural

transitivity, only (154) can be regarded as a typical transitive clause of German. The
same applies to (156) from Finnish. Cases like these pose no problems for our
analysis, since we can justify our claims by referring to semantics. More problematic
in this respect are the following clauses from Hindi and Japanese, cf.

(158) ilaa-ne ek haar(*-ko) uthaayaa

Ila-ERG one necklace.NOM(*-ACC) lift/carry.PERF
‘Ila lifted a necklace’
(159) ilaa-ne haar-ko uthaayaa
Ila-ERG necklace-ACC lift/carry.PERF
‘Ila lifted the/*a necklace’ (Hindi, Mohanan 1994: 63, 79f)

(160) kare ga otoko o korosh-ita

he NOM man ACC kill-PAST
‘He killed the man’

(161) kare wa otoko o korosh-ita

he TOP man ACC kill-PAST
‘He killed the man’
(162) otoko wa kare ga korosh-ita
118

man TOP he NOM kill-PAST

‘He killed the man’
(163) kare wa otoko wa korosh-ita
he TOP man TOP kill-PAST
‘He killed the man’ (Japanese)

Both in Hindi and Japanese, participants in transitive events can be referred to in

different ways. All the examples above profile semantically transitive events, i.e. we
cannot distinguish the different constructions from each other semantically. For
example, in Hindi the nominative (zero) vs. accusative marking of the Patient is
conditioned by definiteness. The event itself is the same in both (158) and (159). In
Japanese, the arguments of transitive clauses can appear in nominative and
accusative cases despite which of them can be marked as topic that is neutral as
regards the semantic role assignment. This is not determined by semantic transitivity
in any way. Which construction in Hindi and which in Japanese should be labelled
as the basic one? As noted, semantics does not aid us in making the decision, since
all the examples designate semantically transitive events. In Hindi, the decision may
be somewhat easier to make, since the differences reflect the typologically relevant
differences in the marking of definite vs. indefinite Patients. Consequently, (159)
should be regarded as more transitive. The marking in Japanese differs from that of
Hindi in that there are two different marking systems operating simultaneously and
independently of each other. These are here labelled as marking of grammatical
roles and topic marking.37 Only the former is related to the expression of transitivity.
Nominative and accusative mark Agent and Patient of transitive clauses
(exclusively). Topic marking, on the other hand, is neutral as regards the semantic
role assignment as well as transitivity of clauses. Any argument or constituent can
be marked as topic. Hence, if we wish to define the basic transitive clause as the
pattern used in the expression of highly transitive events, (160) is our pick for the
basic transitive pattern in Japanese. Also structurally this seems adequate. In this
case, we cannot justify our claim by referring directly to semantics, since the two
marking types do not enable a distinction between transitive and less transitive

37
These labels are ad hoc and are very likely not used in the traditional study of Japanese.
 119

clauses.
Further problems related to the notion of the basic transitive clause are
illustrated by the actual frequency of clauses and the pragmatics of the clauses at
issue. Intuitively, it seems that the most frequently attested patterns should be
labelled as basic (see e.g. Comrie 1988: 19f for differences between passives and
ergative clauses). Above, we have only illustrated cases in which both arguments
are explicitly expressed and marked. However, as has been pointed out by DuBois
(1987) among others typical transitive clauses are very unusual in normal speech in
many languages (see also Siewierska (1997: 189) who states that only 51% of the
237 languages in her sample have case marking on nominal S, A or O38). In a great
number languages, everything that can be inferred from the context is either omitted
or is not marked explicitly. Consequently, many clauses do not have any arguments
at all and cannot be regarded as structurally transitive. It may seem unjustified to
label a clause pattern that is very infrequent in actual language use as the basic
transitive clause of a given language. However, since our definition is based
primarily on semantics, the infrequent occurrence of the given clause type does not
make the definition unjustified. The label ‘typical’ refers here only to the fact that
the clause type is employed in the description of typical transitive events. No matter
how infrequent this pattern is in the actual language use, only it has the potential of
profiling highly transitive events.
As the very brief discussion above shows, we cannot always define the basic
transitive pattern of a given language merely by referring to the semantic transitivity.
Typical transitive events can be expressed using different constructions (including
also derived intransitives). Consequently, we should be able to distinguish between
typical and less typical transitive constructions also in these cases. Since it is not the
goal of this study to propose a universal, completely adequate definition of a typical
transitive clause, the semantic definition outlined above has to suffice. The chief
criteria are the semantic transitivity associated with events and an explicit reference
to both participants. We may perhaps add that the event has to occur in
pragmatically as neutral conditions as possible, i.e. both participants are relevant in

38
Siewierska uses the label P, but we have replaced this by O that is the norm in the present study.
120

the situation described.

4.5.2. On the underlying motivation of transitive marking

In this section, we will discuss the ways in which the marking of basic transitive
clauses is motivated. We start by a general illustration of ‘participant’, ‘core-
participant’ and ‘agent-patient’ marking languages. The last type will be studied in
more detail in sections 4.5.2.2. - 4.5.2.4.

4.5.2.1. ‘Participant’, ‘core-participant’ and ‘agent-patient’ marking languages

As noted above, the transitivity (transitive marking) of clauses may be motivated

primarily by the number of participants, number of core participants (i.e.
participants that inhere in the semantics of the verb) and the semantic transitivity of
events. This can be illustrated in the form of a continuum that is based on the
semantics of the participants involved, cf.

The motivation of transitive marking of clauses:

The number of > The number of core > Semantic roles of

participants participants participants (agent and
patient)

This simple continuum is established on the kinds of restrictions there are for the
use of transitive constructions.39 The kinds of construction at the left end are the
least semantically motivated. The mere number of participants irrespective of their
core or non-core status is a sufficient criterion for structural transitivity. In the
second case, participants must be integral parts (core participants) of events in order
that the employed structure can be transitive. The label ‘core participant’ refers here
to participants that inhere in the semantics of events. For example, in the case of
‘seeing’ both see-er and the stimulus are core participants and required by the

39
Transitive marking refers here to the structure of the construction employed in the description of
typical transitive events (cf. Kozinsky and Lazard in 2.1.).
 121

semantics of the verb, while the location is a non-core participant. Every event is
located somewhere, which is, however, not an integral part of the ontology of events
(even if verbs like inhabit inherently refer to a certain kind of location). In the last
case, the transitivity is more semantically conditioned, since only participants of
highly transitive events allow a transitive marking. The number of relevant factors
grows as we proceed to the right of the continuum. The tripartite division proposed
here is based on clear generalizations, since the transitive marking of clauses is
conditioned by a number of other factors as well. The continuum in its form above
ignores many unclear cases and it only takes account of basic cases. This division
does not aim at classifying languages due to their transitivity marking. This would
require an enormous amount of data and knowledge from a huge number of different
languages. Furthermore, due to space limitations, different motivations are discussed
at a very general level. Hence, each type is illustrated in light of examples from a
single language.
At the extreme left of the continuum, only the number of participants is
relevant and the languages are therefore labelled as ‘participant marking languages’,
cf.

(164) ifá-t pó "si lást-i"-n á "ssi"c-ís

dog-NOM cat black-DUR-OBL chase:LGR-INDIC
‘A dog is chasing a black cat’
(165) cá "ni ‘taló "fa-n hî"c-is
John town-OBL see:FGR-INDIC
‘John sees the town’
(166) cá "ni ‘talo "fa-n ay-áha "n-ís
John town-OBL go:SG-FUT:LGR-INDIC
‘John is going to town’ (Creek, Martin 2000: 380)

Example (164) from Creek describes a transitive event involving two core
participants. Consequently, the marking of the clause is transitive involving a
nominative and an oblique argument. The event profiled in (165) also involves two
participants, but the semantic transitivity is lower than in (164). This is typologically
not unheard of and at first sight (164) and (165) seem like very typical transitive
structures of any nominative-accusative language. What makes the marking
122

interesting, however, is the transitive marking of (166). The event in question

involves two participants in this particular case, but only one of them is a core
participant (semantically), whereas the other (‘to the town’) merely expresses the
goal of a movement and cannot be considered core.
Swedish (along with English) exemplifies a language in which the number of
core participants is the primary factor conditioning the transitive marking of clauses,
cf.40

(167) man-nen måla-de hus-et

man-DEF paint-PAST house-DEF
‘The man painted the house’
(168) man-nen såg stad-en
man-DEF see.PAST city-DEF
‘The man saw the city’
(169) man-nen gå-r till stad-en
man-DEF go-PRES to city-DEF
‘The man goes to the city’

Similarly to Creek, (167) denotes a typical transitive event that causes a clear
change-of-state in the patient. Consequently, the structure employed in (167) can be
regarded as transitive. In (168) the transitivity of the given event is considerably
lower, but the marking remains transitive. On the basis of these two examples
(only), Creek and Swedish are similar as regards the motivation of transitivity.
However, differences arise in (169). In Swedish, the marking is clearly less
transitive in these cases. Hence, the number of core participants (the transitive
valence) is the primary criterion of transitivity. The semantic role inventory makes
a less significant contribution.
The right end of the continuum illustrated above includes cases in which not
only the number, but also the semantic roles of participants contribute to transitivity.
Typically, only events involving an agent and a patient are marked transitively. This
type is here exemplified by Hunzib (Van den Berg 1995: 42ff):

40
There are also cases in which the roles are important as well.
 123

(170) s v-bur-la-l n-uq xo

bear-PL-OBL-ERG 5-eat-PRES meat.5
‘The bears eat meat’
(171) "bu-u ãc’c-r ože
father-DAT 1-see-PRET boy.1
‘Father saw the boy’
(172) b "c’c m-aq’e-r ex-e-™ur
wolf.4 4-come-PRET river-OBL-COM
‘The wolf came to the river’

The event denoted in (170) is highly transitive. Consequently, the clause pattern
employed can justly be considered transitive. Event profiled in (171) displays a
lower degree of transitivity, which is here also reflected in the marking of clauses.
Irrespective of the involvement of two core participants, the marking in (171) is less
transitive than in (170). (172) involves a core participant and a locative phrase. The
marking is clearly less transitive in this case. On the basis of the evidence illustrated
here, we can conclude that Hunzib differs from Creek and Swedish in that the mere
number of core participants is far from being a sufficient criterion for high
transitivity, but the semantic transitivity of the event has to be taken account of as
well. According to Drossard (1991: 413) languages like Creek would be classified
as more transitive than Swedish and Hunzib, since they employ the basic transitive
pattern in the expression of every event involving two participants regardless of the
semantic transitivity of events. Hunzib differs from Creek and Swedish in that in
Hunzib we can infer high transitivity of events directly from the morphosyntactic
form of clauses, since transitive marking is possible only in semantically transitive
cases.
Before proceeding, it must be emphasized that the continuum in the form
illustrated above is based on generalizations about the nature of events and
participants partaking in them. No language is likely to belong purely to one of these
groups, but there are overlaps. For example, in Swedish clauses denoting cognitive
processes, like ‘think’ are expressed by a less transitive construction. However, the
continuum above gives us some insights into the nature of transitivity. It is rather
safe to say that, if language A uses a transitive construction in the marking of any
event involving two participants, it uses a transitive construction also in the marking
124

of transitive events. Transitive marking is not arbitrary in this respect.

Languages like Hunzib differ further from each other in how the transitive
pattern is motivated. The motivation may be primarily semantic or the structure
(syntactic valence) may contribute to the marking as well. In the first case,
arguments are marked according to their semantic roles independently of each other.
In the second case, the syntactic valence is important and the marking is not
individual, but hierarchical (the distinction between these two corresponds largely
to Silverstein’s (1976) local vs. global marking distinction). In the last case, not only
the transitive valence, but the way in which the participants are referred to
contributes to the (in)transitivity of clauses. These are discussed more thoroughly
below. The illustrated types correspond roughly to what Herok (1985: 133f) has
labelled as ‘kumulierende vs. zentrierende Sachverhalts-Präsentation’. According
to Herok, the former of these types represents events as wholes of their parts,
whereas in the latter case one of the participants is semantically ‘core’ (zentral) and
the other is semantically ‘non-core’ (dezentral) (cf. also Silverstein). Despite the
differential motivation, the presented types differ from each other according to
whether the primary motivation is to mark semantic roles of participants or the
overall transitivity of events.

4.5.2.2. Marking of semantic roles

The semantic motivation refers here primarily to the marking of arguments of

transitive clauses according to the semantic roles of their referents. Since the
number of different events is infinite (every event differs both from each other and
from different instances of the same basic event), semantic marking cannot take
account of all differences, but it is based on generalizations. First of all, as noted
above, the explicit marking is in many cases redundant, since we can infer the
semantic transitivity of events and the roles of participants involved from the
semantics of the predicate verb. Consequently, the explicit marking of all possible
differences would be very uneconomical and would burden our processing capacity
enormously. This would also clearly violate the underlying economy principle of
languages (see, e.g. Kibrik 1985: 271). Second, the number of morphosyntactic
 125

mechanisms a language has is clearly outnumbered by the (infinite) number of

different non-linguistic events. This forces languages to make generalizations.
Hence, it must be emphasized that in the following the term ‘semantic role’ always
refers to a generalized role prototype.
In the following, we illustrate two kinds of transitivity marking that can be
said to be primarily conditioned by the semantic roles of the participants involved.
Because of the primary semantic motivation, the structure per se does not impinge
on the marking in any significant way, i.e. the structure and semantics are
independent of each other. Put concretely, this means, for example, that every
argument is marked independently of others. The high transitivity of clauses is
consequently a combination of Agent and Patient marking. Even if the structure as
such is more or less irrelevant as regards the marking of transitivity in the presented
cases, the structure of events (especially their transitive valence) makes a
contribution, as will be shown.
In the first type, the arguments are marked on the basis of the semantic roles
of their referents independently of the transitive valence of events.
Agentive/patientive systems exemplify this, cf.

(173) chi-bashli-li-tok
2.ACC-cut-1.NOM-PAST
‘I cut you’
(174) is-sa-ssa-tok
2.NOM-1.ACC-hit-PAST
‘You hit me’
(175) sa-ttola-tok
1.ACC-fall-PAST
‘I fell’
(176) ittola-li-tok
fall-1.NOM-PAST
‘I fell (on purpose)’ (Choctaw, Davies 1986: 15, 36)

In Choctaw, arguments are marked according to their more agentive or more

patientive nature irrespective of the transitivity of events. Examples (173) and (174)
describe typical transitive events and the structure employed is typically transitive.
126

As can be seen in (175) and (176), the marking is not motivated in any way by the
transitive valence, since both nominative and accusative marking may appear in
(semantically) intransitive clauses as well. That the use of the typical transitive
pattern is restricted to highly transitive events becomes evident, if we take (177)-
(179) into account, cf.

(177) chi-sa-banna-h
2.ACC-1.ACC-want-PRED
‘I want you’
(178) ch «-noksho:pa-li-h
2.DAT-fear-1.NOM -PRED
‘I am afraid of you’
(179) chi-ã-lhakoffi-h
2.ACC-1.DAT-miss-PRED
‘I miss you’ (Davies 1986: 65ff)

Examples above profile events the transitivity of which is considerably lower that
in (173) and (174). This is also reflected in their linguistic marking. In Choctaw, the
transitivity of clauses like (173) and (174) is motivated by the agentive and
patientive features of participants involved in transitive events. The arguments are
marked according to their roles, but the rationale behind the marking is not the
overall transitivity of an event, but the semantic roles of individual participants that
in partake in a transitive event. As regards differences between (173) and (177), for
example, we may conclude that the nominative marking of Agent and the accusative
marking of Patient correlate with high transitivity, whereas any other kind of
marking is associated with a lower degree of transitivity. The transitive marking of
highly transitive events is motivated by the semantic roles of participants on one
hand, and by the high semantic transitivity of events on the other. Arguments are
marked ‘non-nominatively’ and ‘non-accusatively’, if their role is other than typical
agent or patient. The transitive valence of events is irrelevant.
In Choctaw, each argument is marked according to its semantic role
irrespective of the transitive valence. The transitive marking is thus based primarily
on the semantic role of participants individually. The transitive marking is a
combination of agentive and patientive marking of participants in transitive events.
 127

The other type of semantically motivated basic structure is illustrated by languages

in which only arguments of transitive clauses are marked according to their semantic
roles. This means that agents and patients of transitive events are referred to
differently from sole participants of intransitive events even if they bear similar
roles. The crucial difference between these two marking types lies here. In
languages like Choctaw, the sole argument of an intransitive clause is always
marked according to its semantic role. In the other semantically motivated marking
type, the roles are important only if the event in question is transitive. The marking
in intransitive clauses is structurally motivated and is consequently neutral as
regards the semantic role assignment. Pitta-Pitta and Lezgian exemplified below are
examples of languages in which the marking can be claimed to follow primarily this
principle, cf.

(180) õa-tu kat yu-na wat yama-ka kiniyari-õa-na

I-ERG clothes-ACC wash-PAST girl-PURP-ACC
‘I washed the clothes for the girl/the girl's clothes’
(181) yata-ya õan yt ya in-ku
like-PRES I you-DAT
‘I like you’
(182) na õka-ya õan ytya kunti-ina
sit-PRES I house-LOC
‘I’m sitting in the house’
(183) kurra-ka õan yt ya wiri
fall-PAST I LIKE
‘I nearly fell’ (Pitta-Pitta, Blake 1979b: 197, 221)

(184) ada (jak) t’üna

he.ERG (meat) eat.AOR
‘He has eaten (the meat)’ (Lezgian, examples courtesy of Martin Haspelmath)
(185) šarwili.di-z ada-q haj ki ’e xa-na-
Sarwili-DAT he-POSTEL afraid be-AOR-NEG
‘Sarwili was not afraid of him’

(186) balu —-ar wineldi gadar že-zwa

fish-PL upward throw ANTIC-IMPF
‘The fish are jumping up(ward)’
128

(187) zun mad wi q’iliw ata-nwa

I.ABS again you.GEN to come-PERF
‘I have come to you again’ (Lezgian, Haspelmath 1993: 207ff)

The clauses exemplified above describe semantically very different kinds of event.
Only (180) and (184) denote semantically transitive events. In all other cases, the
event is somehow less transitive, which is manifested linguistically. Differences are
due to the explicit marking of different semantic roles. In both Pitta-Pitta and
Lezgian, the ergative marking of Agent implies that the event involves a highly
agentive agent. The accusative or absolutive marking on the Patient, on the other
hand, is related to the affectedness parameter. Consequently, the basic ergative-
accusative/ergative-absolutive marking is restricted to marking of highly transitive
events. In Pitta-Pitta, the Agent has zero marking and the Patient appears in the
dative case, if the event is not typically transitive. This is very natural, since there
is neither agent nor patient present and the marking is primarily related to semantic
roles. The case is very similar in Lezgian where only highly transitive events allow
the ergative-absolutive case frame. Other roles than agent and patient are marked
differently.

4.5.2.3. Marking of grammatical relations (‘syntactically conditioned

transitivity marking’)

Above, we illustrated cases in which the transitive marking of clauses is due to an

explicit marking of semantic roles. In these cases, the arguments were either marked
according to their roles irrespective of the transitivity of events or the explicit
marking was restricted to transitive clauses. The clause structure (primarily syntactic
valence) did not make any major contribution to the marking, since each argument
was assigned its role independently of others. In these cases, both arguments make
an equally important contribution to the transitivity of clauses. In addition to
semantics, the syntactic valence may contribute significantly to the structure of basic
clauses as well. Typical instances of marking types that are more structurally
motivated are illustrated by systems that primarily determined by grammatical
relations. This kind of argument marking is hierarchical in the sense that one of the
 129

arguments is clearly primary, whereas the other(s) is (are) is secondary as regards

its grammatical nature (cf. Herok 1985: 133f). The appearance of the primary
argument is obligatory, i.e. it must appear in every clause irrespective of transitivity.
This has clear consequences for the nature of transitivity marking, as will be shown
below.
In languages like Choctaw or Pitta-Pitta, every argument is marked
independently of others according to its semantic role. This is in accordance with
the semantics of transitive events. Since agents and patients are equally significant
participants in transitive events, both Agent and Patient make a significant
contribution to the transitivity of clauses. Neither argument can be said to be more
important than the other either structurally or semantically. The case is very
different, if the marking is conditioned by grammatical relations. Grammatical
relations are based on a hierarchy in which one argument outranks the other. One
of the arguments is primary and its form is motivated purely structurally. As a
consequence, the marking is completely neutral for the semantic role assignment of
arguments. As for transitivity marking, this means that the marking is less explicit
than in semantically motivated cases. A couple of examples are examined below,
cf.

(188) dí-cca l’urí b-órx-ana

I-ERG rock CL-lift-AOR
‘I lifted up the rock’
(189) dí-ye yas y-ól`-ula
I-DAT girl CL-love-PRES
‘I love the girl’
(190) dun kwaná-na
I.ABS eat-AOR
‘I ate’ (Avar, Lazard 1998: 143, cited from Charachidzé 1981: 154ff)

(191) mies rikko-i maljako-n

man.NOM break-3SG.PAST vase-ACC
‘The man broke the window’
(192) poika rakast-aa äiti-ä-än
boy.NOM love-3SG.PRES mother-PART-3SG.POSS
‘The boy loves his mother’
130

(193) koira kuol-i / hyppä-si

dog.NOM die-3SG.PAST / jump-3SG.PAST
‘The dog died/jumped’ (Finnish)

In languages like Avar and Finnish, only either argument is marked explicitly. The
other (primary) argument is always unmarked. Depending on the language, the
marked argument of transitive clauses is either Agent (Avar) or Patient (Finnish).
Examples above describe very different events. As can be seen, the typical transitive
marking is also here restricted to highly transitive events, i.e. the marking is partly
semantically motivated as well. However, the marking in Finnish and Avar differs
from that in Choctaw, Pitta-Pitta and Lezgian in that only the marked
(grammatically secondary) argument reflects changes in transitivity. The primary
argument retains its marrking despite obvious changes in transitive valence and/or
semantic transitivity. This is due to the structural motivation of the primary
argument. Consequently, we may say that languages like Avar and Finnish primarily
mark the overall transitivity of events rather than the semantic roles of individual
participants. It would not be justified to say that the transitive marking in Avar is
based solely on the explicit expression of agency, whereas in Finnish the marking
is based only on the explicit expression of affectedness. This is not to say that these
languages do not emphasize these facets. The transitive marking of clauses is only
not due to the marking of agency or affectedness, but to the notion of high
transitivity in general. In similar vein, the different marking of the secondary
(semantically marked) argument in (189) and (192) is responsible for marking all
the changes in the semantics of the events denoted. In order that we could justly
claim that the marking is motivated only by the given aspects, agency should always
correlate with changes in the affectedness and vice versa. Due to the unmarkedness
of the primary argument, the ergative marking of the Agent or the accusative
marking of the Patient is alone responsible for marking transitivity. Hence, these
markers imply high transitivity and cannot appear in intransitive clauses.

4.5.2.4. Reflection of transitivity

So far, we have been illustrating cases in which the transitivity of clauses has been
 131

motivated directly by the transitivity of single events denoted. The transitive

marking may also follow from transitivity of complex events. In other words, the
transitive marking is not necessarily motivated by the high transitivity of a single
clause, but a transitively marked clause may be a part of a ‘transitive network of
events’. As regards the motivation of the marking, this kind of marking illustrates
a kind of subgroup of the latter of the types illustrated above, since the marking
emphasizes the high overall transitivity. A couple of examples of this kind of
marking are illustrated and discussed below, cf.

(194) w Ç misa-n  dugu sya-e tel-

the woman-ERG goat kill-INF start-PERF
‘The woman started to butcher the goat’
(195) w Ç misa(-n ) (ja) na-e tel-
the woman(-ERG) (rice) eat-INF start-PERF
‘The woman started to eat (rice)’
(196) w Ç manu den-e tel-a
the man(ABS) sleep-INF start-PERF
‘The man began to (go to) sleep’ (Newari, Givón 1985: 96ff)

(197) na-mpi v-kwalca-tay- cut

3SG O-3DL A-rise-try-RM PAST
‘They both tried to get him up’
(198) na-mp v-taray-wapat- cut
3SG O-3DL A-clear-ascend-RM PAST
‘Those two let him loose and went up’ (Yimas, Foley 1991: 333, 340)

(199) niy-ala wad’jay

he-first go 3SG RA-PAST
‘He went first’
(200) ngindaji gurama gayga’wuma winthali
this-ERG man chop FUT-3SG<3SG-ma2 firewood
‘This man has to cut firewood’

(201) niy-ingga wad’jay

the.former-ERG come 3SG-ra-PAST
‘He (previously mentioned) came’ (Bunuba, Rumsey 2000: 104ff)
132

In the exemplified cases, transitivity is not directly motivated by the transitivity of

the ‘main verb’, but follows rather from the overall transitivity of the verb complex.
The presented cases are somewhat different. In Newari, the obligatoriness of the
ergative marking correlates with the semantic transitivity of the overall event.
Examples (194)-(196) profile events in which an agent starts to perform an action.
The actions themselves differ crucially from each other in transitivity. The act of
beginning per se is neutral for transitivity. We can start performing intransitive or
transitive events. These differences are reflected morphosyntactically in Newari. The
ergative marking is obligatory in (194), since the action to be performed is highly
transitive. In (195), the event is clearly transitive, even if it may be regarded as
somewhat less transitive than in (194). This results in the optionality of the ergative
marking. The changes in the marking are clearly motivated by the overall
transitivity, since only events displaying very high transitivity obligatorily trigger
the ergative marking. Example (196) profiles an intransitive event. Consequently,
the ergative marker is omitted. The transitive vs. intransitive marking of clauses is
due to the degree of transitivity of the ‘secondary’ event. The examples from Yimas
are a kind of mirror image of those from Newari. The verbs kwalca and wapat are
monovalent verbs that only allow one argument to be explicitly expressed. The
expression of two arguments (i.e. the transitivity of the clauses) is made possible by
bivalent tay and taray. The transitivity of one of the verbs is a sufficient criterion for
transitive marking. Examples from Bunuba illustrate transitive marking that is even
more clearly motivated by the transitivity of the overall transitivity of a complex
sequence of events. Example (199) exemplifies the typical intransitive pattern of
Bunuba. The ergative marking of Agent is in Bunuba typically motivated by the
transitivity of the clause, as can be seen in (200). Example (201) illustrates a
peculiar marking (also according to Rumsey, see 2000: 108), since the sole
argument of an intransitive clause is marked ergatively. The marking does not
follow from agency, since for this to be the case the marking should appear in (199)
as well. Instead, the marking is due to the transitivity of the overall event. Example
(201) is taken from a text about an emu that is going to the river to eat. The going
illustrated in (201) is a part of a complex transitive event with a definite goal. Thus,
(201) receives a transitive encoding despite the clear intransitivity of the event itself.
 133

4.6. A structural typology of the basic transitive clause

4.6.1. Theoretical background

As illustrated and discussed in the previous sections, the transitive marking of

clauses can be motivated in one of two ways: the transitivity may be primarily due
to the marking of semantic roles or the marking may be associated with the high
overall transitivity of events. Furthermore (closely related to the motivation of the
marking), languages can be divided into four according to which aspect(s) of high
transitivity they focus on. Highly transitive events involve an agent and a patient and
languages can emphasize none, either or both of these explicitly. Typical examples
of the first type are analytic languages that lack morphological means of marking
arguments explicitly. In these cases, a language has no ‘option’ but to mark
transitivity implicitly. There are, however, languages more synthetic in their nature
that also express transitivity only explicitly (as will be illustrated). The second
possible type, in which only either of the facets is explicitly marked, is nicely
illustrated by typical accusative and ergative systems. The former emphasizes the
affectedness parameter and the latter the agency parameter. The last type is
illustrated by the type that is motivated by the semantic roles of participants
involved. Before illustrating the types, it must be emphasized that differently from
the previous section the distinction is here based solely on the ‘outcome’ of the
motivations illustrated above. In other words, the different underlying motivations
of the marking are ignored and the typology presented is based merely on the aspect
marked explicitly. Each of the types is illustrated and discussed below, cf.

(202) monna o-tlilê

man SAG-came
‘The man came’
(203) monna o-rôbilê selêpê
man SAG-broke axe
‘The man broke the/an axe’ (Northern Sotho, Louwrens et al 1995: 44ff)

(204) Hasan öl-dü

Hasan.NOM die-PAST
134

‘Hasan died’
(205) müdür mektub-u imzala-d x
director.NOM letter-ACC sign-PAST
‘The director signed the letter’ (Turkish, Comrie 1989: 175f)

(206) dhibi bara-wa-nha

bird.NOM rise-MOVING-PRES
‘The bird is flying’
(207) bura:-dhu gabuga: ga-miyi
child-ERG egg.ABS break-TR.PAST
‘The child broke an egg’ (Ngiyambaa, Donaldson 1980: 190, 220)

(208) háama hipáayna

man 3.NOM.came
‘The man came’
(209) háana-nm pée-‘wiye wewúkiye-ne
man-ERG 3ERG-shot elk-DO
‘The man shot an elk’ (Nez Perce, Rude 1988: 547f)

Northern Sotho exemplifies a language that does not mark any of the relevant facets
of high transitivity explicitly. There is nothing in the form of either argument or in
the verb morphology that could be regarded as an explicit transitivity marker. The
only relevant feature is the number of arguments. The mere number of arguments
is here not regarded as an explicit marker of transitivity, since also less transitive
events have in many cases two arguments none of which is an agent or a patient.
Turkish is a typical example of a language that marks the affectedness parameter
explicitly in transitive clauses. The marking is not related to the affectedness
parameter per se, since if this was the case, the marking should appear in (204) as
well. In Ngiyambaa, the only explicitly marked transitivity parameter is agency. Nez
Perce is an exceptional language, since in transitive clauses both the Agent and the
Patient are marked according to their semantic roles (cf. Pitta-Pitta above). Hence,
both aspects of high transitivity are marked explicitly on the relevant arguments
themselves.
The differences in marking exemplified above do not correlate with any major
semantic differences between the events in question. This means that all transitive
 135

clauses illustrated above are equally transitive. To our knowledge, there is no

evidence to support the fact that speakers of predominantly accusative and
predominantly ergative languages would observe events differently as regards their
transitivity (cf. e.g. Herok 1985: 146 and Dixon 1994: 214ff). The marking types
exemplified above merely illustrate different structural possibilities of describing
transitive events. However, even if the semantic differences between the illustrated
types are practically non-existent, the differential emphasis has consequences for the
nature of transitivity alternations encountered. An illustrative example of this
provided by the appearance of morphologically distinct passives in accusative
systems and similar antipassives in ergative systems. The analogous functions (i.e.
‘antipassives’ in accusative and ‘passives’ in ergative systems) are usually expressed
simply by omitting the Patient (‘antipassive’) or the Agent (‘passive’). Typical
examples are cited below, cf.

(210) der mann ass (den brei)

ART.NOM man eat.PAST.3SG (ART.ACC porridge)
‘The man ate the porridge’
(211) der brei wurde gegessen
ART.NOM porridge become.PAST.3SG eat.PARTIC
‘The porridge was eaten’ (German)

(212) (im-u-di) hamaXi ’inni

(father-OBL-ERG) donkey beat.PAST
‘The father beat the donkey/the donkey got beaten’
(213) ¨ali-di q’iru b=el-at-a-da
Ali-ERG wheat N=thresh-PRES-CONV-COP
‘Ali is threshing wheat’
(214) ¨ali w=ol-a-da
Ali M=thresh-AP.CONV-COP
‘Ali is threshing’ (Godoberi, Kibrik 1996: 110,137)

Both German and Godoberi have the means to express both passive and antipassive
functions. The languages differ from each other in what kind of consequences for
the structure of the clause the expression of the function has. In German (along with
most accusative languages), the passivization is marked on the verb, whereas
136

antipassivization is not. Furthermore, the marking of the sole argument in passive

changes from accusative to nominative. The case is converse in Godoberi: the
passive is structurally similar to the German antipassive and antipassive to the
German passive. This has nothing to do with obviously different semantics of the
functions in the exemplified languages, but the primary reason here lies in the
semantic transitivity parameter emphasized.
Above, languages were subdivided due to the aspect(s) of semantic transitivity
they emphasize. Below, we proceed to proposing a more detailed typology of the
basic transitive clause. The term ‘basic transitive clause’ is here used in accordance
with the definition given by Kozinsky and Lazard (cf. 2.4.), i.e. as a basic transitive
clause is here defined every clause that is identical with the construction that is
employed in the description of highly transitive events. The typology is based on
comparisons of transitive and intransitive clauses. In a given language, any
structural feature that is restricted to transitive clauses is viewed as a transitivity
marker. The features taken into account in the typology proposed include verb
morphology, verb agreement and the number and marking of arguments. We will
only take account of the direct marking and ignore the possible indirect marking.
Furthermore, only basic transitive clauses are relevant, derived transitives (along
with ditransitives) are ignored. However, as the alert reader may notice derived
transitives will be used as examples in some cases. In these cases, the marking of
the verb is not regarded as a feature of transitivity marking, but only verb agreement
and argument marking are taken account of. There are two reasons to ignore derived
transitives. First, derived clauses are not basic, but they are always somehow
marked. As regards the transitivity of clauses, this refers mainly to differences in the
verb morphology. In some cases we may be led to think that the explicit marking of
the verb is an integral part of transitivity in a given language. Because of this, we
always try to exemplify transitive and intransitive clauses that cannot be derived
from each other (suppletive cases are accepted, e.g. cases like ‘kill’ vs ‘die’).
Second, apart from the verb morphology, the differences between basic and derived
transitive constructions are rather minor in many languages. Taking account of
derived transitives would not make any significant contribution to our typology. For
example, in the following, the only difference between basic and derived clauses is
 137

the verb morphology, cf.

(215) bayi ya a ina- u

ABS man.ABS sit-NFUT
‘The man is sitting down’
(216) balan `ugumbil ba õgul yara-õgu inay-ma-n
ABS woman.ABS ERG man.ERG sit-APPL-NFUT
‘The man is sitting down with the woman’
(217) balan `ugumbil ba õgul ya a-õgu balga-n
ABS woman.ABS ERG man-ERG hit-NFUT
‘The man is hitting the woman’ (Dyirbal, Dixon 1972: 96, 58)

Example (215) illustrates the typical intransitive pattern of Dyirbal, whereas (216)
is a derived transitive of it. (217), on the other hand, illustrates the basic transitive
clause of the given language. As can be seen, the only difference between derived
and basic transitive clauses is the marked verb morphology in (216). However,
considering (216) as the basic clause of Dyirbal is not justified. Marked verb
morphology is not a basic property of every transitive clause in Dyirbal and
consequently cases like (216) are ignored here. Ditransitives are not taken account
of, since they do not make any significant contribution to the marking of basic
transitivity. Ditransitive clauses do not denote events that would somehow be
significantly more transitive than basic transitive ones (cf. below). Furthermore, they
do not include relevant structural features that would not be parts of basic transitive
clauses (this is not to say that they are completely irrelevant as regards transitivity,
but they are not relevant for the purposes of this section).
The most fundamental and basic difference between intransitive and transitive
clauses is the number of arguments. Every language must have some
morphosyntactic mechanism at its disposal to express this difference. The simplest
way of expressing this difference is simply to refer to both arguments without
specifying their semantic roles more closely. Moreover, the marking of (either or
both) arguments may be affected and the verb morphology and agreement may also
be different. The number of arguments involved also contributes to the complexity
of clauses: intransitive clauses are less complex than the transitive clauses. As for
the features of transitivity, this means that transitive clauses have more relevant
138

features than intransitive ones, i.e. transitivity is marked explicitly. Languages in

which intransitivity is marked more explicitly are not common. This also means that
the relevant transitivity features are typically features of transitive clauses only. In
the simplest case, the only such feature is the number of explicitly expressed
arguments (this is the case in all instances of quantitative type). The appearance of
a certain feature alone suffices to distinguish transitive clauses from intransitive
ones. For example, in most accusative languages accusative Patients can appear only
in transitive clauses. In order that a mechanism can be regarded as a marker of
transitivity, it has to be attested only in transitive clauses. This means, for example,
that not the mere marking of arguments is a sufficient criterion, but the
marker/mechanism in question has to make it possible to distinguish transitive
clauses from intransitive ones. For example, in the following examples from
Pitjantjatjara (Bowe 1990: 10), only the ergative marking can be regarded as a
marker of transitivity41, cf.

(218) billy-nya a-nu

Billy-ABS go-PAST
‘Billy went’
(219) tjitji-ngku billy-nya nya-ngu
child-ERG Billy-ABS see-PAST
‘The child saw Billy’

If we only take account of example (219), we might be (mis)led to think that the
absolutive affix -nya is an explicit Patient marker (with proper nouns) and can thus
be considered a transitivity marker. However, this affix is functionally an absolutive
marker, since it also appears in intransitive clauses. Hence, it does not qualify as a
genuine transitivity marker. Below, some mechanisms of this kind will be taken
account of. They are, however, not regarded as genuine transitivity markers (this is
usually explicitly stated), but they may still contribute to the typology of basic
clauses.
In principle, there are two possible kinds of transitivity marker. First, there are

41
The ‘explicit’ absolutive marking is restricted to proper nouns.
 139

markers that can only appear in transitive clauses. For example, the ergative affix
in (219) exemplifies this. In these cases, transitive clauses are structurally more
complex than intransitive ones. The other possibility is that the intransitive clause
has an explicit marker that is omitted in transitive clauses. This kind of marker is at
best a theoretical possibility and I have not been able to find any example of this.
Difficulties in finding an example of this kind of marker are easily understood, if we
take account of the ‘complexity principle’ noted above. Transitive events involve
two arguments, which usually produces a more complex structure in other respects
as well. Hence, it would be odd to find cases in which the intransitive clauses are
somehow more marked than transitive ones. The following examples from
imaginary English illustrate how this marker may look like, cf.

(220) *he-m come-s

3SG-NOM come.PRES-3SG
(For: he comes)
(221) *he hit-s I
3SG.ABS hit-3SG.PRES 1SG.ABS
(For: he hits me)

In this case, the ‘nominative marker’ -m would be an explicit marker of

intransitivity. Since the appearance of this marker is restricted to intransitive
clauses, the zero marking of either (or both) argument(s) would be an implicit
marker of transitivity. The absence of a marker in intransitive clauses is not regarded
as an explicit transitivity marker here.
In the following presentation, we use the term ‘added argument’ to refer to the
argument that is introduced in the transitive clause. In similar vein, the argument
that is structurally identical to the only argument of an intransitive clause is referred
to as ‘original argument’. The adding does not have to coincide with semantics, but
the argument that structurally differs from S, is always regarded as added, even if
it refers to the same participant as the S argument. Thus, in (222) and (223), the
accusatively marked Patient is regarded as the added argument, cf.

(222) mies kuol-i

man.NOM die-3SG.PAST
140

‘The man died’

(223) susi tappo-i miehe-n
wolf.NOM die.PAST-3SG man-ACC
‘The wolf killed the man’ (Finnish)

Example (223) represents a lexical causative and we are dealing with a case in
which an agent is added to an originally intransitive event. However, the
accusatively marked Patient is here viewed as the added argument, since it deviates
structurally from S. This kind of use of the term agrees with the structural emphasis
of our typology. Even if we define the basic transitive pattern primarily by referring
to semantics, we are more concerned with the structure below. All cited examples
do not profile highly transitive events. This is, however, irrelevant, as long as it is
obvious that we really are dealing with the basic transitive pattern. Cases in which
the lower semantic transitivity is directly reflected in the marking of clauses are
naturally excluded.
Since the proposed typology here is based on differences between intransitive
and transitive clauses, it is also in order to define what we mean by a basic
intransitive clause before going into details. Basic intransitive clause is here defined
as a clause pattern that is employed in the description of events involving one (if
possible, concrete) participant that is performing an intransitive action (excluding
reflexives) or that is undergoing an event that is out of its control. Dynamicity is
important here, since in many languages clauses denoting states are structurally
untypical intransitive clauses. For example, in Finnish and German (along with
many other languages) stative non-actions are expressed differently from dynamic
intransitive events, cf.

(224) minu-a palel-taa

I-PART be.cold-3SG.PRES
‘I am cold’
(225) talo on suuri
house.NOM COP.PRES big
‘The house is big’
(226) jaakko kontta-a / kaatu-u
jacob.NOM crawl.PRES-3SG / fall.down-3SG.PAST
 141

‘Jaakko is crawling/falling down’ (Finnish)

(227) mich frier-t

I.ACC be.cold.PRES-3SG
‘I am cold’
(228) das haus is-t gross
ART.NOM.SG house.NOM COP.PRES-3SG big
‘The house is big’
(229) der mann läuf-t / stirb-t
ART.NOM.SG man.NOM run.PRES-3SG / die.PRES-3SG
‘The man runs/dies’ (German)

Examples (224)-(225) and (227)-(228) illustrate non-typical intransitive patterns of

Finnish and German, whereas (226) and (229) are examples of the basic pattern.
Choosing the latter structures as the basis of our typology is in line with our
definition of basic transitive event. Basic transitive events are dynamic two-
participant events in which one of the participants is an agent and the other a patient.
Events denoted in (226) and (229) are semantically much closer to basic transitive
events than mere states, since they involve either of the relevant aspects of typical
transitive events (i.e. an agent or a patient). In Finnish and German, this is reflected
in the clause structure. If we base our typology on differences between transitive
clauses and stative intransitives, we may be emphasizing irrelevant features. For
example, the verb agreement in both (224) and (227) is restricted to third person,
whereas the verb agrees in person and number in typical intransitive clauses in both
languages. These kinds of difference are attested in many other languages as well.
The most radical case is that these clauses are verbless. Consequently, if we choose
clauses like (224) or (227) as the starting point of our study, we may be led to think
that verbal agreement is a feature of transitive clauses only. In the typology below,
we always aim at using basic intransitive clauses (as defined here) as the point of
comparison to transitive ones. This might, however, not be possible in every case
and less typical examples may be used as well. Real problems arise only in cases
like those illustrated above. In some languages, typical and less typical intransitive
clauses do not differ from each other in the way illustrated above.
We begin our presentation by a general distinction between marking on the
142

verb and marking on the arguments. Languages can be divided into two according
to which of these two they use as the primary marker of transitivity. Some languages
use both of these mechanisms. After this, we illustrate a more detailed typology of
the basic transitive clause.

4.6.2. Verb- and argument marking languages

Languages can be divided into three according to whether they mark basic
transitivity primarily on the verb, on the arguments or on both. The division is made
on the basis of the primary morphosyntactic location of the marking. This means,
for example, that verbal marking covers here also the marking of the arguments on
the verb by person affixes, even if this can also be regarded as an instance of
argument marking. The division here is based only on the expression of basic
transitivity, i.e. derived transitive constructions are excluded. The labels employed
here coincide with ‘head-marking’ (verb-marking) and ‘dependent marking’
(argument marking) introduced by Johanna Nichols (see e.g. Nichols 1986). We
prefer the terms ‘verb-‘ and ‘argument-marking’, since the term ‘verb-marking’
covers here two different kinds of marking (see below). Furthermore, there are cases
in which neither of these terms is applicable (see e.g. (243)-(246)).
In order that a language can be regarded as a primarily verb marking language,
arguments have to be marked neutrally as regards the transitivity of the denoted
event. Transitive and intransitive clauses cannot be distinguished on the basis of
argument marking, even if the number can, of course, be different. All the relevant
transitivity features are expressed on the verb. There are (at least) two different
kinds of verb marking languages that differ from each other in the nature of what is
signalled on the verb. There are languages that primarily mark the verb according
to the transitivity of clauses and languages that mark arguments on the verb
according to their semantic roles (in this case the arguments themselves are neutral
in this respect). The former type is illustrated by (Boumaa) Fijian and the latter by
Abaza, cf.

(230) e kama a+i-sulu

3SG burn garment
 143

‘The garment is burning’

(231) e kama-ca a+i-sulu a cauravou
3SG burn-TR garment the youth
‘The youth is burning the garment’ (Boumaa Fijian, Dixon 1988: 47)

(232) a-ph <cs d-qa-c <’a-d

DEF-woman 3-hither-sit-PAST(ACT)
‘The woman sat up’
(233) a-ph <cs a-qac’a d-l-š c-d
DEF-woman DEF-man 3-3F-kill-PAST(ACT)
‘The woman killed the man’ (Abaza, Anderson 1976: 4)

In (Boumaa) Fijian, the basic transitivity of clauses is expressed primarily with

verbal affixes. The -(C)a-affix appears on transitive verbs, but not on intransitive
ones. Abaza exemplifies a language in which arguments are distinguished solely in
the verb. In (232) and (233,) arguments are marked on the basis of their definiteness,
but neutrally as regards their semantic roles. The marking (i.e. cross-reference) on
the verb is in this case alone responsible for transitivity marking.
Despite the primarily verb marking languages, there are languages in which
the marking of arguments is primarily (or solely) responsible for the marking of
transitivity. As opposed to the cases illustrated above, the verb behaves neutrally as
regards the marking of transitivity, cf.

(234) ngali-mpa ngunytju a-nu

1DU-PURP mother go-PAST
‘Our mother went’
(235) minyma-ngku ngayu-nya pu-ngu
42
woman-ERG 1SG-ACC hit-PAST
‘The woman hit me’ (Pitjantjatjara, Bowe 1990: 16, 24)

(236) äiti-mme men-i

mother-1PL.POSS go-3SG.PAST
‘Our mother went’

42
Note that in (218)-(219) the same morpheme -nya is glossed as ABS.
144

(237) äiti-mme tappo-i kissa-n

mother-1PL.POSS kill-PAST-3SG dog-ACC
‘Our mother killed a/the cat’ (Finnish)

In Pitjantjatjara, the verb does not agree in any way (person, number, gender etc.)
with either argument irrespective of transitivity of clauses. This means that the case
marking on arguments is the sole indicator of transitivity. The verb expresses
transitivity also indirectly, since the form of the past affixes is sensitive to
transitivity. In Finnish, the verb always agrees with the nominative subject, i.e. one
of the arguments is marked on the verb. However, because the cross-referenced
argument is always the same regardless of transitivity, we do not view the verb
agreement as a transitivity marker. It does not amount to distinguishing transitive
clauses from intransitive ones. The sole indicator of transitivity is the accusative
marking of the Patient.
The third type of language marks transitivity both on the verb and on the
arguments. Examples of this kind of language are not hard to find, cf.

(238) er patck neme ra ptc-p§e cnqen qora-nta-l§-c-n

next.day again go.home-TH DEM .3SG.ABS reindeer.herd.PARTIC-E-3SG.ABS
‘The next day that herder again went home’
(239) q cnwer jara-lp-c-jõ-c-n õew-§ctt§-c-qe-e
finally house-SG-E-AUG-3SG.ABS woman-dog-E-DIM-ERG
tejk-c-nin
make-E-3SGA.3SGO
‘(Then they moved camp), finally the bitch made a big house’ (Chukchi, Dunn 1999: 80f)

(240) õatju ka-na pula-mi

1SG.ABS PRES-1SG.NOM shout-NONPAST
‘I am shouting, I shout’
(241) õatjulu-lu ka-na-õku njuntu njanji
1SG-ERG PRES-1SG.NOM-2SG.ACC 2SG.ABS see.NONPAST
‘I see you’
(242) njuntulu-lu ø-npa-tju pantu-nu õatju
2SG-ERG Ø-2SG.NOM-1SG.ACC spear-PAST 1SG.ABS
‘You speared me’ (W arlpiri, Hale 1973: 309, 328, 336)
 145

Both in Chukchi and Warlpiri, the argument marking follows a double marking
pattern. This means that nominal arguments are marked according to an absolutive-
ergative pattern, in addition to which the arguments are cross-referenced by
agreement affixes. Both marking systems mark transitivity independently of the
other and would consequently sufficiently mark clauses either as intransitive or
transitive. However, there is nothing optional about the marking, but transitivity is
obligatorily signalled by both relevant elements in Chukchi and Warlpiri.

4.6.3. The typology

In the following, we proceed to illustrating a typology of the basic transitive clause.

As noted above, the typology is based on explicit differences between intransitive
and transitive clauses. This way we gain the relevant information about the
expression of transitivity. A mere typology of transitive clauses without any
reference to the intransitive constructions of the languages in question would not
reveal the very aspects we are most interested in. We will distinguish four major
clause types (that all include many subtypes). The distinction is based on the
marking of arguments and verb. In the first type, the primary difference between
intransitive and transitive clauses is the number of arguments. There are usually no
other differences involved. Type 2 includes cases in which there is no marking on
the arguments themselves, but the verb is responsible for indicating transitivity (and
intransitivity) of clauses. The marking on the verb covers also cases in which the
arguments are marked on the verb by person affixes. It is important that the
independent arguments are not marked in any way. The chief structural feature of
type 3 is the explicit marking of either argument. We finish the typology by
illustrating cases in which both arguments are marked. The presentation follows a
‘cumulative’ principle. This means that the relevant features of type 1 are
automatically a part of type 2, features of type 2 are included in the description in
type 3 etc. Put more concretely this means, for example, that the cross-reference of
arguments on the verb that is a relevant feature of type 2 (but not one of type 1) is
also taken account of in the description of type 3. Consequently, the number of
possible types increases as we proceed.
146

4.6.3.1. Type 1

Since the most fundamental difference between intransitive and transitive clauses
is the number of arguments, the simplest type comprises cases in which this is the
only difference between intransitive and transitive clauses.43 The added argument
is structurally identical to the S argument. Single features are not alone sufficient
criteria of transitivity, but the whole clause has to be taken into account when
defining transitivity. This kind of transitivity marking is here referred to as
quantitative. Teribe and Yoruba are illustrative examples of this types, cf.

(243) carlos shro-no kupke

Carlos arrive-PERF yesterday
‘Carlos arrived yesterday’
(244) carlos jacinto shpo-no
Carlos Jacinto hit-PERF
‘Carlos hit Jacinto’ (Teribe, Quesada 2000: 107)

(245) omo dé
child(ren) come
‘The children came’
(246) omo rà aso
child(ren) buy fabric
‘The children bought fabric’ (Yoruba, Itkonen 1997: 161)

The only difference between intransitive and transitive clauses is the number of
arguments. Neither of the arguments is in any way marked according to its semantic
role or the transitivity of the described event. Furthermore, there are no changes in
the verbal agreement, since the verb is in both languages neutral in this respect.
In Teribe and Yoruba, transitive clauses are formed by ‘reduplicating’ the S
argument. The two arguments are structurally identical. The basic transitive clauses
of the languages illustrated above exemplify what is here labelled as ‘symmetric’
marking system. As for basic transitive clauses, this means that transitive clauses

43
The constituent order is ignored in the present context, even if it is the only way of distinguishing
between Agent and Patient, i.e. A O V and O A V are not regarded as distinct types here.
 147

have two arguments, both of which are marked in the same way. The status of the
arguments is also very similar. As a result, it is rather difficult to say, which of the
arguments is primarily responsible for the expression of transitivity. Moreover, the
verb agreement is neutral in Teribe and Yoruba. The other symmetric type is
illustrated below, cf.

(247) kuñ-t korh-wë-t

house-3F sitting-IMPF-3SF
‘It is a house’
(248) yima-r noh-më-r
man-3SM die.REM.PAST-3SM
‘A man died’
(249) yima-r yën-t gëbrërna-më-r-t
man-3SM child-3SF rub-REM.PAST-3SM-3SF
‘The man rubbed a girl’ (Alambak, Bruce 1984: 182ff)

Examples from Alambak are identical to those from Teribe and Yoruba with the
crucial difference that the verb agrees with the core arguments irrespective of
transitivity. In intransitive clauses, the agreement is with the sole argument, whereas
in transitive clauses, both arguments are cross-referenced on the verb. What
amounts to labelling this marking system as symmetric lies in the form of the
agreement affixes. Structurally, the affixes employed in intransitive and transitive
clauses are identical. Hence, transitive clauses are formed by adding an argument
to the intransitive clause and by cross-referencing the argument in the verb (note that
the order of affixes matches with that of arguments). As in Teribe and Yoruba, there
are no morphological indicators of transitivity, but the differences between
intransitive and transitive clauses are purely quantitative.
The symmetry of marking in Teribe, Yoruba and Alambak is based on the
equal status of the arguments in transitive clauses. As for the morphosyntactic
status, both arguments of transitive clauses are identical to the S argument. The
other, asymmetric, quantitative type is illustrated by the following examples from
Iatmul (Foley 1986: 102f) and Ejagham (Watters 2000: 213)
148

(250) ntvw y v-ntv

man go-3SG.M
‘The man went’
(251) ntvw takw c v v-ntv
man woman see-3SG.M
‘The man saw the woman’

(252) èbh 0n ábh4 é-rîk

C5.farm 3PL.POSS C5.PFV-burn
‘Their farm burned’
(253) yê à-ríg-ì èbh 0n ábh 4
C1.3SG 3SG.PFV-burn-CAUS C5.farm 3PL.POSS
‘He burned their farm’

The main difference between intransitive and transitive clauses in Iatmul and
Ejagham is also the number of arguments. However, there is one further difference
that makes it justified to view the basic transitive clause of these languages as
slightly different from the symmetric type illustrated above. This is the obligatory
S/A agreement of the verb. As can be seen, the verb always agrees with the subject
both in intransitive and transitive clauses in both exemplified languages. In order for
the agreement to be symmetric the added (Patient) would also have to agree with the
verb. The structural status of the added argument is thus a bit lower than in Teribe,
Yoruba and Alambak.
The differences between intransitive and transitive clauses can be quantitative
and/or qualitative. The former of these types refers, for example, to the mere number
of arguments (or person affixes): intransitive clauses have one argument and
transitive ones have two. In the case of qualitative markers, some structural feature
associated with the marking is restricted to transitive clauses only. Hence, there is
something in transitive clauses that is not a part of the ‘corresponding’ intransitive
clause. The given marker distinguishes transitive clauses from intransitive ones
explicitly. In some languages, arguments themselves are marked neutrally for
transitivity, but transitivity is indicated by some other mechanism. An example of
this is illustrated below, cf.
 149

(254) i-crer
1-sing
‘I sang’
(255) a-jõt
2-sleep
‘You slept’
(256) i-te a-pupun
1-PAST 2-see
‘I saw you’
(257) a-te ih-kre
2-PAST 3-plant
‘You planted it’ (Canela-Krahô, Popjes & Popjes 1986: 129ff)

As can be seen above, arguments are marked (or referred to) in the same way both
in intransitive and transitive clauses. The main difference is also here the number
of arguments. As opposed to Teribe and Yoruba, however, there is a minor
difference that can be claimed to directly indicate differences in transitivity. This is
the appearance of the past tense marker -te that is restricted transitive clauses. If the
Agent is a pronoun, the marker is very closely attached to the pronoun in question,
as can be seen. Cases like these illustrate a kind of indirect marking, so we are not
dealing with a genuine transitivity marking, but the status of the marker is somewhat
lower.
In cases illustrated thus far, there is nothing in the form of the arguments or
in the cross-reference of the verb in transitive clauses per se that would
unambiguously distinguish them from intransitive ones. Both arguments could
because of their form appear also in intransitive clauses. On the basis of this, the
following examples from Lakhota and Haida also exemplify this type, cf.

(258) Ø-wa-kté
3SGU-1SGA-kill
‘I killed him’
(259) ma-Ø-kté
1SGU-3SGA-kill
‘He killed me’
150

(260) wa-hí
1SGA-arrive
‘I arrived’
(261) ma-khúže
1SGU-sick
‘I am sick’ (Lakhota, Van Valin 1985: 365f)

(262) »k’in xa giyu »ā qayd-ān

woods DIST towards-FOREGROUND 1SG.AG go-PAST
‘I went up into the woods’
(263) h āwnu di xwi-g ān
there 1SG.PAT cold-PAST
‘I was cold there’
(264) la »ā qiõ-gig ān
3SG.PAT 1SG.AGT see-PAST
‘I already saw him’
(265) di la squda-g ān
1SG.PAT 3SG.AGT hit-PAST
‘He hit me’ (Haida, Mithun 1999: 214, cited from Levine 1977)

Lakhota and Haida both exemplify a rather typical agent/patient system attested in
different forms in many languages of the Native North America. In these systems,
the transitive marking of clauses is motivated by the explicit semantic role
assignment of the participants involved (cf. 4.5.2.3. above). As for the structural
marking of transitivity, this means that also in these cases the number of arguments
is the sole indicator of transitivity. Differently from Teribe, Yoruba and Alambak,
the arguments or cross-reference affixes in transitive clauses are not identical in
form, which is due to the motivation of the marking. What amounts to viewing these
systems as subtypes of the quantitative type is the lack of any individual, qualitative
feature that would only be found in transitive clauses. Transitive clauses are in
languages like Lakhota and Haida formed by combining two intransitive (an
agentive and a patientive one) into one. Furthermore, because of the underlying
semantic motivation of the marking neither of the arguments can be regarded as a
primary one.
 151

4.6.3.2. Type 2

In the previous section, we illustrated primarily quantitative cases in which there is

nothing in transitive clauses that could not as such be attested in intransitive clauses
as well. In this section, we proceed to examining cases in which the differences
between intransitive and transitive clauses are more obvious, since transitivity is
also signalled on the verb (qualitatively). The illustrated structural features include
verb agreement (or cross-reference) and verb morphology. Similarly to type 1, there
are no differences in the marking of independent arguments.
Above, we already illustrated cases in which intransitive and transitive clauses
diverge as regards the verb agreement. For example, in Alambak the verb agrees
with both arguments in transitive clauses. However, we are not dealing with a
qualitative marking of transitivity in these kinds of case, since both arguments in
transitive clauses are cross-referenced by the same marker that is used for the sole
argument of intransitive clauses. In the following, we will focus on cases in which
the verb agreement is also clearly an indicator of transitivity, since certain features
are restricted to transitive clauses. The simplest instance of this is illustrated by
cases in which the agreement as a whole is restricted to transitive clauses. Two
different examples of this are represented below, cf.

(266) e ije ruo

man the come
‘The man came’
(267) e ije barone
man the die
‘The man died’
(268) e ije ame kan-ia
man the child hit-3PLU
‘The man hit the children’
(269) e ije ame kan-i
man the child hit-3SGU
‘The man hit the child’ (Barai, Foley 1986: 104f, cited from Olson 1981)

(270) ùbúr á-túuk’

Ubur CMPL-play
152

‘Ubur played’
(271) ùbúr jòobì á-kèel-é
Ubur buffalo CMPL-shoot-3SGA
‘Ubur shot the buffalo’ (Päri, Andersen 1988: 292f) 44

Clauses (268) and (269) seem very similar to those from Iatmul with the difference
that in Barai the Patient is cross-referenced instead of the Agent. However, if we
compare intransitive clauses of Barai and Päri to those from Iatmul obvious
differences arise. In (266), (267) and (270), the agreement is neutral, since in Barai
and Päri the verb agreement is restricted to transitive clauses. Consequently,
differently from Iatmul, the agreement markers in (268), (269) and (271) can be
regarded as genuine transitivity markers. They distinguish transitive clauses
explicitly from intransitive ones. The Patient (Barai) or the Agent (Päri) agreement
of the verb is the only indicator of transitivity (along with the number of arguments
naturally) in the examples illustrated above.
In Barai and Päri, the mere agreement suffices to distinguish between
transitive and intransitive clauses. Another, perhaps more widely attested type, is
illustrated by cases in which the distinguishing factor is not the mere agreement, but
the number of affixes along with their form is different, cf.

(272) mtoto a-na-soma

child 3SG-PRES-read
‘The child is reading’

(273) mtoto a-li-mw-ona mwalimu

child 3SG-PAST-3SG.OBJ-see teacher
‘The child saw the teacher’ (Swahili)

(274) na aì na-ta.mbuta dàngu amung

ART wood 3SG.NOM-drop.out with root
‘The tree is uprooted’
(275) na tau wútu na-palu-ka nyungga
ART person be.fat 3SG.NOM-hit-1SG.ACC 1SG

44
The latter is an alternate construction to an ergative one.
 153

‘The big man hit me’ (Kambera, Klamer 1998: 63)

In Swahili and Kambera, core arguments are cross-referenced on the verb. The form
of independent arguments is insensitive to transitivity, since both arguments of
transitive clauses are identical to the S argument. What separates the marking
system of Swahili and Kambera from that of Alambak, for example, is the form of
the agreement affixes. In Alambak, the cross-reference is symmetric, i.e. the form
of the affixes is exactly the same both in intransitive and transitive clauses. In
Swahili and Kambera, on the other hand, the agreement is organized
asymmetrically. One of the affixes is structurally identical to the affix that is used
in intransitive clauses, whereas the introduced affix differs structurally from the S
agreement affix. This affix is also the one that is most directly responsible for the
marking of transitivity, since it only appears in transitive clauses.
We can rather safely claim that in languages like Swahili and Kambera, the
cross-reference of arguments is the primary indicator of transitivity. First, as noted
above, the marking of arguments functions on a symmetric basis, i.e. independent
arguments are not marked according to their semantic roles in transitive clauses.
Second, and perhaps more importantly, in languages like these, the independent
arguments are very often eliminated and participants are referred to by agreement
affixes only. The expression of arguments is (structurally) optional, while person
affixes are obligatory. This is possible, since the cross-reference on the verb suffices
it to distinguish between agents and patients. This also distinguishes transitive
clauses from intransitive ones.
Barai, Päri, Swahili and Kambera illustrate languages in which the agreement
per se is a direct indicator of transitivity. Additionally, the sole form of agreement
affixes can indicate (in)transitivity. Chamorro exemplifies this kind of system, cf.

(276) h-um-anao yo
1SG-SG-go 1SG.ABS
‘I went away’
(277) hu-sakke i guihan
1SG.ERG-steal the fish
‘I stole the fish’
154

(278) mu-lalalo’ si tata-hu

SING-mad UNM father-1SG.POSS
‘My father became mad’
(279) ha-li’e si juan si maria
3SG.ERG-see UNM John UNM Maria
‘John saw Mary’ (Chamorro, Cooreman 1987: 37ff)

In (276) and (277), there is an independent argument and an agreement affix on the
verb. The difference between these two examples is found in the nature of the verb
agreement. In (276), we are dealing with the same kind of verb agreement as, for
example, in Iatmul. The prefix h-um- refers to the absolutive pronoun yo. On the
other hand, in (277), the prefix hu- does not agree with the noun ‘fish’, but it refers
to the agent (that is not expressed as an independent pronoun). The nature of the
agreement is here clearly different. Based on these two examples, we might claim
that in intransitive clauses the verb prefix is an agreement marker (in the ‘Iatmul
sense’), whereas in transitive ones it cross-references the Agent that is not expressed
as an independent argument. However, the picture is not this simple, as indicated
in (278) and (279) in which participants are referred to by independent arguments.
Because of the intransitivity of the event, (278) has one argument, whereas transitive
(279) has two. The verb agrees in (278) (by default) with the only expressed
argument. In (279), the verb agrees with the Agent. At first sight, this system seems
identical to that of Iatmul. The crucial difference is here found in the form of the
agreement prefix. The form of the prefix is sensitive to transitivity. These agreement
types are labelled as absolutive and ergative agreements by Cooreman (1987: 36).
However, it seems that we are here dealing with a tripartite marking, since S, A and
O all behave differently as regards the verb agreement. S is cross-referenced by the
prefix mu- in (278), the prefix for A is ha-, whereas O is not cross-referenced at all.
Since the S agreement is restricted to intransitive clauses and A agreement to
transitive clauses, the mere form of the agreement affix suffices to distinguish these
clauses from each other. Also in Chamorro, the (in)transitivity is indicated solely by
the agreement affixes.
In (266)-(279), we have illustrated cases in which the marking on the verb has
exclusively referred to the marking of arguments. This can be regarded as a kind of
 155

argument marking that appears on the verb instead of the nominal arguments
themselves. In addition, there are languages in which we are rather dealing with
morphological marking of transitivity. The function of the verbal marking is not to
cross-reference arguments. It is in order to emphasize that causative or applicative
marking on the verb is excluded from the description below. Causatives and
applicatives present derived transitives and the morphological marking of the verb
is in this case a part of the alternations in question and it is thus not related to the
marking of basic transitivity. A couple of examples of genuine transitivity marking
on the verb are illustrated below, cf.

(280) para tamu masuk bersama(-sama)

PL guest enter together
‘The guests entered together’
(281) ia mem-buka pintu ini
s/he meN-open door this
‘S/he opened this door’ (Bahasa Indonesia, Varamasi 1999: 10, 30)

(282) luma kera qe nii karangi

house 3PL 3SG.NFUT be.located be.close
‘Their house is close’
(283) wane e laba-tani-a maka nau
man 3SG.NFUT harm-TRANS-3SG.OBJ father 1SG
‘The man harmed my father’ (Toqabaqita, examples courtesy of Frank Lichtenberk)

(284) ni §ím cš |cN s |éni§

AUX walk ART.woman
‘The woman walked’

(285) ni q’wc-ct-cs ›cN s |éni§ tccN scé "|tcn

AUX bake-TRANS-3.TR.SUBJ ART.woman ART.salmon
‘The woman baked the salmon’ (Halkomelem Salish, Kroeber 1999: 38)

In Bahasa Indonesia, transitivity is signalled only by manipulating the verb

morphology. The verb does not agree with either argument and there is no case
marking of any kind on the arguments. Transitive verbs are usually marked with
meN- in typical transitive clauses, whereas intransitive verbs are unmarked. The
156

marking in Toqabaqita and Halkomelem Salish is somewhat different. Identically

to Bahasa Indonesia, the transitivity is morphologically signalled on the verb and
furthermore there is no marking on the nominal arguments. What distinguishes these
languages is that the Patient in Toqabaqita and the Agent in Salish is cross-
referenced on the verb, whereas the verb of Bahasa Indonesia is neutral in this
respect.

4.6.3.3. Type 3

Up to now, we have been illustrating cases in which transitivity has been primarily
marked on the verb (the number of arguments is taken for granted). In what follows,
we will examine cases in which transitivity is also signalled on (pro)nominal
arguments. Since the added argument is marked differently from the original one,
the marking is always asymmetric. These systems have already been illustrated in
passing in light of Swahili and Kambera. In these cases, the arguments are explicitly
distinguished, but the morphosyntactic ‘location’ of the marking is different from
that at issue in this section. As a consequence, the morphological marking on the
arguments themselves is here viewed as distinct from argument marking on the
verb. As regards the separation of arguments, these two systems can be regarded as
two varieties of the same basic system.
In the simplest case, intransitive and transitive clauses are distinguished only
by marking the added argument differently from S. The marking of the original
argument remains the same. A couple of typical examples are illustrated below, cf.

(286) ma õkula pu õka-na

child fall-PAST
‘A child fell down’
(287) ma õkula talku-na yukuru-ku
child strike-PAST dog-ACC/DAT
‘A child struck a dog’
(288) yukuru pilya-na ma õkula-ku
dog bite-PAST child-ACC/DAT
‘A dog bit a child’ (Ngarluma, Blake 1977: 7f, cited from Hale 1968)
 157

(289) õia waka-mu

I fall-PAST
‘I fell’
(290) õa-tu kupi wala-mu
I-ERG fish kill-PAST
‘I killed a fish’
(291) kupi-õku õia taca-mu
fish-ERG I bite-PAST
‘A fish bit me’ (Yalarnnga, Blake 1977: 8)

Ngarluma illustrates a very typical nominative-accusative marking system. This

means that the only explicitly marked argument in transitive clauses is the Patient.
Since there is no marking on the verb, the accusative marking can be regarded as the
sole indicator of transitivity in this case. The basic transitive clause of Yalarnnga is
the exact mirror image of the corresponding structure of Ngarluma, since the Agent
is the only explicitly marked argument in transitive clauses. Also in Yalarnnga, the
ergative marking of the Agent is the only marker of transitivity.
In Ngarluma and Yalarrnga, the somewhat lower morphosyntactic status of the
added argument is reflected only in case marking. It may be reflected in other ways
as well. One such feature is the verb agreement or cross-reference. The agreement
of the added argument usually differs from that of the original one. In many
languages, only one of the arguments is cross-referenced on the verb and it is in
many cases the argument that is morphosyntactically identical to the sole argument.
This is attested e.g. in Finnish, Berber and Archi, cf.

(292) mies hyppä-si

man.NOM jump-3SG.PAST
‘The man jumped’
(293) jaakko raken-si torni-n
Jacob.NOM build-3SG.PAST tower-ACC
‘Jacob built a tower’
(294) salamurhaaja ampu-i miehe-n
assassin.NOM shoot-3SG.PAST man-ACC
‘The assassin shot the man’ (Finnish)
158

(295) t-rzem tewwurt

3FS-open door.CST
‘The door opened’
(296) y-rzem wryaz tawwurt
3MS-open man.CST door
‘The man opened the door’
(297) y-bedd wrba
3MS-stand boy.CST
‘The boy is standing’
(298) y-wtu wrba aqqzin
3MS-hit.PERF boy.CST dog
‘The boy hit the dog’ (Berber, Guerssel 1986: 22, 11, 9)

(299) buwa d-akdi

mother.CLASS II.ABS CLASS II-go.away
‘Mother went away’
(300) dija w-akdi
father.CLASS I.ABS CLASS I-go.away
‘Father went away’
(301) buwa-mu dija o-w-ka
mother.CLASS II-ERG father.CLASS I.ABS bring-CLASS I
‘Mother brought father’ (Archi, Kibrik 1997: 281f)

The basic transitive clauses of Finnish, Berber and Archi differ from those
illustrated in (286)-(291) because of the obligatory verb agreement with the
unmarked (nominative or absolutive) argument. This might be claimed to underline
the lower status of the added argument. However, since the verb agreement is the
same in both illustrated patterns, it cannot be regarded as a genuine marker of
transitivity. Consequently, also in these cases, the only marker of transitivity is the
accusative or ergative marking of the added argument. The agreement does not
distinguish intransitive clauses from transitive ones. However, since the lower status
of the added argument is also reflected in the verb agreement, it is in order to
distinguish this clause from that illustrated above. Examples from Finnish and
Berber differ from each other in that in Finnish the nominative is clearly a zero
marked argument, whereas Berber is a language with a marked nominative. Despite
the more explicit marking of the nominative, it is functionally identical to a typical
 159

zero marked nominative. We are not dealing with any kind of ergative marking here,
even if the Agent is seemingly the marked argument in transitive clauses.
In all the cases exemplified so far, the marking has been directly related to the
independent arguments themselves. A structurally somewhat different type of a
marking system in which one of the arguments is marked differently is illustrated
in the following examples from Wari45, cf.

(302) mao na
go 3SG.AG&REAL
‘(S)he went’
(303) to’ na-on (>non)
hit 3SG.AG&REAL-3SG.PAT
‘(S)he hit him’
(304) mao na piye’
go 3SG.AG&REAL child
‘The child went’
(305) to’ na-m narima tarama
hit 3SG.AG&REAL-3SGF.PAT woman man
‘The man hit the woman’

As the examples represented above show, neither the marking of arguments nor the
verb morphology distinguishes between transitive and intransitive clauses. However,
we are justified to regard the marking system of Wari as an instance of the clause
type illustrated in this section, because the form of the person clitic is sensitive to
transitivity. The employed mechanism is simply different. In (302) and (303),
participants are referred to only by the person clitic. In (302), the clitic refers to one
participant only and the form is na. In (303), a (masculine) Patient marker is
attached to the Agent marker, because the event at issue is transitive. Examples
(304) and (305) illustrate that there is no marking on independent arguments
themselves. Because of the unmarkedness of both the verb and the arguments, Wari
clearly belong to type 1. However, regarding this kind of system as subtype of type

45
All the examples from W ari are taken from handouts given by professor Esa Itkonen in his class
‘Introduction to Amazonian languages’ held at University of Turku, fall 2001. The primary source
is ‘W ari’ by Dan Everett and Barbara Kern.
160

1 does not do justice to the nature of the system in question. In Wari, the added
argument is explicitly marked differently from the original one as regards the
marking on the clitic element. The very ‘location’ of the marking just happens to be
different from that we are more familiar with.
In Finnish, Berber and Archi, the cross-reference of the added argument is
different from that of S. However, because we are dealing with zero marking, the
agreement cannot be regarded as any kind of genuine transitivity marker. There
would have to be obvious differences for this to be the case. Below, we will
illustrate examples from languages in which the added (pro)nominal argument, as
well as the agreement affix are different from that used in the intransitive clause, cf.

(306) la:-Ø si-ke

leopard-ABS die-PFV
‘The leopard died’
(307) tip clkya-e la:-Ø s cih-ke-o
Tipalkya-ERG leopard-ABS kill-PFV-3SG
‘Tipalkya killed a leopard’ (Kham, W atters 1988: 139)

(308) omo-de alme mon-i...

big-ATTR shaman(NOM) say-3SG.INTR
‘The great shaman said...’
(309) met es’ie tet pulut-kele kudede-m
my father(NOM) your husband-ACC kill-3SG.TR
‘My father has killed your husband’ (Kolyma Yukaghir, Maslova 1998: 111f)

(310) ~mu che=ku on-a

3SG house=LOC go-3SG.PAST
‘He went to the house’
(311) ~mu-n lokhu ton-ju
3SG-ERG water drink-3SG.PAST
‘He drank water’ (Dolakh ~ New ~r, examples courtesy of Carol Genetti, LSA 2001)

(312) angut-Ø sana-vuq

angut-ABS work-3SG.INTR.IND
‘The man is working’
(313) qimmi(q)-up arnaq-Ø kii-vanga
dog-ERG woman-ABS bite-3SG/3SG.TR.IND
 161

‘The dog bit the woman’ (Inuktitut, Nowak 1996: 107, 79)

(314) alä-Ñ kulo-Ñ

man-DEF die-3SG.PRET
‘The man died’
(315) ava-Ñ šav c-ïc a µä-Ù
woman-DEF kill-3SG.3SG.PRET man-GEN.DEF
‘The woman killed the man’ (Mordvinian, examples courtesy of Nobufumi Inaba)

In all these examples, the added argument is marked distinctively from S. The S
argument is always zero-marked, whereas the added argument appears either in the
accusative or in the ergative case. In addition to the differences in marking, the verb
agreement varies according to transitivity of clauses. In Kham, the agreement seems
to be restricted to transitive clauses in which the verb agrees in person and number
with the Agent. In Kolyma Yukaghir, the verb agrees in both intransitive and
transitive clauses with the nominative argument. Basically, the agreement follows
the same pattern, as for example in Finnish, Archi, Berber, Iatmul and Ejagham.
However, differently from these languages, the form of the agreement affix is
sensitive to transitivity. In intransitive clauses, the affix is -i, while in transitive
clauses the affix changes to -m. We are here dealing with a genuine transitivity
marker that clearly distinguishes transitive clauses from intransitive ones. The affix
-m cannot appear in intransitive constructions. The primary difference between
Kolyma Yukaghir and Dolakh~ New~r is that in Kolyma Yukaghir, the marking
pattern in nominative-accusative, whereas in Dolakh~ New~r the marking follows
an absolutive-ergative pattern. Moreover, the verb agreement changes in Dolakh~
New~r: in intransitive clauses the agreement is with the absolutive argument and in
transitive clauses with the morphologically marked ergative argument. The status
of the absolutive argument is affected, too. Inuktitut and Mordvinian differ from the
other languages under study here in that the number of arguments cross-referenced
on the verb changes. In intransitive clauses, the verb agrees (by default) with the
only explicitly expressed argument (and participant in the described event), whereas
in transitive clauses, both arguments are cross-referenced on the verb. However, the
marking is here regarded as a subtype of patterns like those of Dolakh~ New~r and
Kolyma Yukaghir. This is due to the morphological nature of the agreement
162

markers. As opposed to Dolakh~ New~r and Kolyma Yukaghir, both arguments are
cross-referenced by suppletive marker that cannot (synchronically) be divided into
(agentive and patientive) components meaningfully.
Examples (292)-(315) illustrate cases in which the verb agreement follows a
constant pattern in the sense that the agreement is always either the same as in
intransitive clauses or the agreement shifts from the unmarked to the marked (i.e.
added) argument. We may conclude that in the former case, only primary
(unmarked) arguments can be cross-referenced on the verb, whereas in the latter
case the agreement is neutral in this respect. There are also languages in which the
picture is much fuzzier, since the marking in transitive clauses may vary. An
example is illustrated below, cf.

(316) m § aal-o
1SG-Ø came-1SG
‘I came’ (Marathi, Klaiman 1987: 73)
(317) ti keel khaa-t-e
3SG.F banana eat-PRES-3SG.F
‘She eats a banana’
(318) ti ni keli khaa-ll-it
3SG.F ERG banana-PL eat-PERF-3PL
‘She ate bananas’
(319) ti ravi laa chal-l-a
3SG.F Ravi ACC torture-PRES-3SG.F
‘She tortures Ravi’
(320) ti ni ravi laa chal-l-a
3SG.F ERG Ravi ACC torture-PERF-NEUT
‘She tortured Ravi’ (Marathi, Blake 1994: 130, cited from Rosen & W ali 1988: 5)

As predicted, the verb agrees with the unmarked S in intransitive clauses. In

transitive clauses, the marking varies according to the markedness of arguments. If
both arguments are unmarked for their roles, the verb agrees with the Agent, as in
(317). On the basis of (316) and (317), we might conclude that the agreement
follows a nominative-accusative pattern. However, this is violated in (318) where
the verb agrees with the zero marked Patient. In (319), the agreement of the verb is
with the Agent again. In (318) and (319), the agreement pattern is quite
 163

straightforward, since the verb always agrees with the zero marked argument. As
shown in (320), the verb cannot agree with a marked argument, but the agreement
is neutralized. In Marathi, the agreement of the verb does not follow a constant
principle, but the marking is either with the unmarked argument or with the Agent,
if both arguments appear in a zero marked form. Consequently, a straightforward
comparison of intransitive and transitive clauses is not possible as regards the verb
agreement.
In Kham, Dolakh~ New~r, Kolyma Yukaghir and also Marathi, the verb agrees
both in intransitive and transitive clauses only with one argument (the semantic
nature of which may vary). Intransitive and transitive agreement differ from each
other either in the absence vs. presence of the agreement (Kham), the form of the
marker (Kolyma Yukaghir) and/or in the nature of the argument that is cross-
referenced (Dolakh~ New~r). In Inuktitut and Mordvinian, the agreement markers
refer to both arguments, but it is structurally impossible to tell the components apart.
Moreover, there are languages in which the added (nominal) argument differs
structurally from the typical S argument, in addition to which the arguments are both
indexed on the verb by affixes, one of which is different from the agreement marker
used in intransitive clauses, cf.

(321) ma chin b’eet-a

ASP 1SG.ABS walk-1SG
‘I walked’
(322) ma Ø-b’eet-a
ASP 2SG.ABS-walk-2SG
‘You walked’
(323) ma tz’-ok n-tzeeq’a-n-a
ASP 2SG.ABS-DIR 1SG.ERG-hit-DS-1SG/2SG
‘I hit you’ (Mam, Palmer 1994: 62f, cited from England 1983)

(324) bärhe särihu

Berhe work.PAST.3SG.MASC
‘Berhe worked’
(325) bärhe n c-‘astir harimu-wa
Berhe ANIM-Astir hit.3SG.MASC-3SG.FEM
‘Berhe hit Astir (woman’s name)’
164

(326) ‘astir n c-bärhe harima-tto

Astir ANIM-Berhe hit.3SG.FEM-3SG.MASC
‘Astir hit Berhe’ (Tigrinya, Palmer 1994: 232, 7f)

(327) yinami gayurba yani-õa-la yinagu

this woman go-PRES-3SG(S/A) here
‘This woman is coming here’
(328) gunami-õgu bama-õgu bandu-li-la-õga gunami gayurba
that-ERG man-ERG hit-PAST-3SG(S/A)-3SG(P) that woman
‘That man hit that woman’ (Biri, Blake 1977: 2, cited from Beale 1974)

(329) gia be-ge ragasi

3PL REM-3PL arrive
‘They arrived’
(330) gia-na bai be-ge gani-a
3PL-ERG pig REM-3PL eat-3SG
‘They ate the pig’ (Sinaugoro, Tauberschimdt 1999: 2)

In Mam, Tigrinya, Biri and Sinaugoro the transitivity is marked both on the nominal
arguments and through the cross-reference of the verb. In Mam the marking on
nominals and the agreement of the verb both follow an absolutive-ergative pattern,
while in Tigrinya both of these mechanisms are organized on a NOM-ACC basis.
Biri and Sinaugoro, on the other hand, illustrate a kind of ‘split system’ in which the
marking on the nominals follows an absolutive-ergative pattern, while the cross-
reference is nominative-accusative. As a consequence, the exemplified languages
differ from each other in how the added argument is cross-referenced on the verb
in comparison with S. In Mam the added argument is marked differently from S. In
addition, the affix cross-referencing the added argument is different from that used
for S. The marking of the added argument is entirely responsible for the marking of
transitivity, since ergativity is restricted to transitive clauses. In Tigrinya the verb
of intransitive clauses agrees with S that is not marked for case. The marking of the
introduced argument (O) is different, in addition to which O is also cross-referenced
by a different marker from that used for A (that agrees with the verb in the same way
as S). Biri and Sinaugoro differ from Mam and Tigrinya in that the marking cannot
be regarded as symmetric in the sense that the marking in Mam and Tigrinya can.
 165

Biri and Sinaugoro can both be regarded as predominantly absolutive-ergative as

regards the marking of independent arguments. In this respect, transitive clauses
differ from intransitive ones in the same way as, for example, in Yalarnnga and
Dolakh~ New~r. What distinguishes Biri and Sinaugoro from these languages is the
differential verb agreement. In Biri, there are, in principle, two independent marking
patterns both of which mark transitivity but differ from each other in what facet of
transitivity they emphasize. As noted, the marking on the nominal arguments is
organized on an absolutive-ergative basis. In this case the explicit marking on the
Agent is responsible for the marking of transitivity. As for the verb agreement, on
the other hand, transitivity is signalled by the Patient agreement marker that appears
only in transitive clauses. The status of the added Agent argument is in Biri more
problematic than in Mam. The nominal Agent ranks lower in status than the zero
marked Patient. However, the status may be said to somewhat higher than for
example, in Archi (cf. above), since the Agent is cross-referenced on the verb by the
same agreement affix as S in intransitive clauses (that can be regarded as the
primary argument). In this respect, O is the explicitly marked argument. Systems
like this are an intermediate form between languages like Yalarnnga and Roviana
(cf. below). The former of these languages mark transitivity only by marking one of
the arguments of transitive clauses differently from S, whereas in the latter case both
are marked explicitly. In Biri, the marking on nominals emphasizes one aspect and
the verb agreement the other. The basic transitive clause of Sinaugoro is very similar
to that of Biri. The main difference is found in the nature of the cross-reference. As
can be seen in (329), S is cross-referenced in the auxiliary verb (that in this case
expresses tense). In transitive clauses, the A is cross-referenced by the same
agreement affix as S. Differently from S, the Agent is marked by the ergative marker
-na. The Patient is not marked for case and it is structurally identical to S. However,
the Patient is cross-referenced on the verb differently from S. The S argument is
cross-referenced in the auxiliary verb, whereas the unmarked Patient is cross-
referenced in the main (semantic) verb. It cannot be inferred from the examples
cited, whether there is a formal difference between the markers (i.e. whether -a is
a genuine Patient marker), but since the affixes appear on different elements, they
can be regarded as distinct markers. Also in Sinaugoro, the marking on the nominals
166

and the verbal cross-reference disagree. The former emphasizes agency, while the
latter signals the patient.
In Biri and Sinaugoro, transitive and intransitive clauses are distinguished
based on three different criteria. First, the marking of the added argument is
different. S and O arguments are zero marked, whereas the Agent appears in the
ergative case. Second, the unmarked O is cross-referenced on the verb by a cross-
reference affix different from that used for S in intransitive clauses. Third, the
original cross-reference shifts to the added argument. The mechanism itself is not
introduced, but the nature of the argument in question has changed from a zero
marked to a marked argument. In these cases, two of the markers are such that they
explicitly distinguish transitive clauses from intransitive ones. Similar cases are
attested also in (one subsystem of) Roviana and in Ngan’gityemerri, as illustrated
below. Before saying anything about the examples above it is in order to state that
the clauses exemplified here seem to be rather marginal in Ngan’gityemerri.
Participants seem to be rather rarely referred to by independent arguments. The
examples used are among the very few cited by Reid that involve lexical arguments.
The claims made here are based on evidence from these examples only, cf.

(331) taloa se zima

leave ABS Zima
‘Zima left’
(332) seke-i-a e maepeza se zima
hit-TR-3SG.DO ERG Maepeza ABS Zima
‘Maepeza hit Zima’ (Roviana, Corston 1996: 12f)

(333) kuru di-m-purrngpurrng nyine

water 3SG.S-Sit-boil FOC
‘The water is boiling now’
(334) kuru ngi-nem-Ø-purrngpurrng
water 1SGA-Heat-3SG.O-boil
‘I’m boiling the water’
(335) wa-mumu-ninggi de-m-burr-tiptip (wurrun)
M-taboo-AGT 3SG.A-Hands-3PL.O-pick up (3PL)
‘The policeman arrested them’ (Ngan’gityemerri, Reid 2000: 342ff)
 167

In both languages, the distinction between intransitive and transitive clauses is based
on three different criteria all of which are as such sufficient indicators of transitivity.
The added argument is marked differently from S, the verb is structurally (and/or
semantically) transitive and the cross-reference of the verb in transitive clauses is
somehow different from the intransitive agreement. The employed mechanisms are
also somewhat different. In Roviana, transitivity is explicitly expressed by the
ergative marking of the Agent, the appearance of the transitive affix -i on the verb
and Patient agreement of the verb. All of these are restricted to transitive clauses.
Roviana is similar to Biri and Sinaugoro in that the status of both arguments of
transitive clauses differs from that of S. The Agent appears in a different case form,
whereas the (absolutively marked) Patient is cross-referenced on the verb only in
transitive clauses. In addition to differences in the status of arguments a transitive
affix is attached to the verb. The transitivity marking in Ngan’gityemerri is in
principle identical to Roviana, since also in Ngan’gityemerri three different
mechanisms are employed. The nature of these mechanisms is, however, somewhat
different. As can be seen, the marking of lexical arguments is organized on an
ergative basis in Ngan’gityemerri, i.e. Agent is marked differently from O and S
arguments. An Agent marker is attached to the Agent, even if the Patient is not
expressed (in (335) the Patient is optional). The cross-reference follows a strict
nominative-accusative pattern (cf. Reid 2000:334). This is also the case in Roviana.
The two languages differ from each other in that in Roviana the accusativity
manifests itself in the fact that the verb agrees only with the Patient, whereas in
Ngan’gityemerri both arguments are cross-referenced on the verb. The other, much
more interesting difference between Roviana and Ngan’gityemerri lies in the way
the verb is marked as transitive. In Roviana, the transitivity of the verb is signalled
by the affix -i. It is structurally rather obvious that (332) and not (331) illustrates the
transitive pattern. In Ngan’gityemerri, on the other hand, the transitivity of
predicates is marked by intransitive vs. transitive finite verbs. For example, in (333)
the finite verb is intransitive ‘Sit’ and in (334) transitive ‘Heat’. The finite verbs
have to match the transitivity of the lexical main verb that is intransitive in (333)
and transitive in (334)-(335). Deviations are ungrammatical. Examples (333)-(334)
illustrate the use of these verbs as markers of transitivity alternations. The lexical
168

verb is the same in both cases, but the transitivity (i.e. the number of participants
partaking in the event) varies. In the latter kind of case, the finite verb is primarily
responsible for the (in)transitivity of clauses. One of the most important functions
of these verbs is to make the marking of arguments on the verb possible. In
Ngan’gityemerri, the person affixes cannot be attached directly to the lexical verbs,
but they only adhere to finite verbs. This results in an explicit marking of both
intransitive and transitive verbs in Ngan’gityemerri.

4.6.3.4. Type 4

Most of the cases illustrated up to now have exemplified clause types in which only
one of the arguments in transitive clauses is marked distinctively from S. Examples
from Biri, Sinaugoro and Roviana gave some foretaste of transitive clauses in which
the marking of neither argument of transitive clauses is fully identical to S. In Biri,
Sinaugoro and Roviana, the marking of the added nominal argument is different
from the unmarked O (or S), whereas the verb agreement varies according to
transitivity. Hence, the status of neither argument is identical to that of S. A further
type of this kind of marking is illustrated by languages in which both nominal
arguments are marked differently from S. This refers to languages with tripartite
marking patterns. These are rare typologically, but not unheard of. The main reason
for this nmight inhere in the clear violation of the underlying economy principle of
language use (cf. e.g. Kibrik 1985: 271). This kind of system is very redundant as
regards the disambiguation of semantic role assignment which is the primary
function of argument marking (see e.g. Blake & Mallinson 1981: 51). As a
consequence, this kind of system is attested only in a handful of languages some of
which are exemplified below, cf.

(336) kiniyari nan-pa-ka kanti-ma-ka

girl she-NOM-here go-AROUND-PAST
‘The girl went for a walk...’ (Pitta-Pitta, Blake 1979b :224)
(337) pukarra-na tatjia matjumpa-lu
grass-OBJ eat.PRES roo-ERG
‘The kangaroo is eating the grass’
 169

(338) õamari-lu takunu-na watjamaia

mother-ERG child-ACC wash.PRES
‘Mother is washing the baby’ (Pitta-Pitta, Blake & Breen 1971: 87, 90)

(339) kana-ia palu-õa

man-NOM die-PAST
‘The man died’
(340) kana-ulu kalka-õa titi-nana
man-ERG hit-PAST dog-ACC.FEM
‘The man hit the bitch’ (W angkumarra, Blake & Mallinson 1981: 51)

(341) háama hipáayna

man 3.NOM.came
‘The man came’
(342) háana-nm pée-‘wiye wewúkiye-ne
man-ERG 3ERG-shot elk-DO
‘The man shot an elk’ (Nez Perce, Rude 1988: 547f)

(343) m ~ hal§ huyase

1SG.FEM.NOM went.SG.FEM AUX.PAST.1SG.FEM
‘I (feminine) had gone’
(344) m ã chokaria kh ‘ m ~rio
1SG.OBL girl.OBL OBJ beat.null concord
‘I beat the girl’ (Sindhi, Klaiman 1987: 72)

In three of the exemplified languages (Wangkumarra is the exception), the S

argument appears in a zero form, whereas both the Agent and the Patient are marked
explicitly in transitive clauses. In Wangkumarra, on the other hand, all of the
arguments (including nominative S) are explicitly marked. There is no
morphologically zero form. The nature of the differential marking of both arguments
is different from that in Biri, Sinaugoro and Roviana, since the differences are
signalled on independent arguments instead of being divided between independent
arguments and the predicate verb. Moreover, the exemplified languages differ from
each other in whether and how the ‘transitivization’ affects the verb agreement. In
Pitta-Pitta and Wangkumarra, the predicate verb does not agree with any argument
irrespective of transitivity. In Nez Perce, the verb agreement follows a nominative-
170

accusative pattern, since in (341) the verb agrees (by default) with S, whereas in
(342) it shows agreement with the Agent (the agreement marker is glossed as
ergative). This kind of agreement pattern is quite consequent, because there is no
unmarked argument in the transitive clause. Since both arguments are marked in
transitive clauses, the agreement must be with a marked argument. In Nez Perce, the
given argument is the Agent (S/A agreement seems to be more typical cross-
linguistically). As a result of the differences in verbal agreement, the agreement
would by itself be a sufficient indicator of transitivity. Sindhi illustrates a rather
interesting language type as regards the verbal agreement. In this respect, Sindhi
seems to follow the same pattern as Marathi (cf. above). In intransitive clauses, the
verb agrees with the nominatively marked S, as is exemplified in (343). In (344),
both arguments appear in an oblique form, in addition to which the Patient is
marked with the Patient particle kh‘. As a result, the verb shows null concord, i.e.
it agrees neither with Agent nor with Patient. As in Pitta-Pitta and Nez Perce, the
primary indicator of transitivity is clearly the explicit marking of both arguments.
However, the null concord of the verb may also be claimed to signal transitivity,
since the null concord is restricted to transitive clauses. Consequently, Sindhi
illustrates a marginal language type in which the intransitive clause is in some
respects more marked than the transitive one.

4.6.4. Summary and discussion

We have illustrated a rather detailed typology of the basic transitive clause.

Different types have been illustrated in light of concrete examples from a variety of
languages. In this section, we briefly sum up some of the most significant points
made above. We also present the proposed typology in a schematic, more abstract
form in order to illustrate an overview of the findings of the previous section and to
give the reader an easier access to the typology.
As illustrated above, the most important morphosyntactic mechanisms that
distinguish transitive clauses from intransitive ones are the number of arguments
(which is a prerequisite for transitive marking), the marking of arguments and the
verb morphology (both cross-reference and other morphological changes). The
 171

number of arguments is the most straightforward reflection of transitivity: clauses

that involve one argument are intransitive and those with two are transitive. This
feature can be explained by referring to non-linguistic evidence only. Other aspects
of transitive marking are linguistic. The picture is clouded by the marking of
arguments, since not every instance of explicit argument marking can be regarded
as transitive, but some patterns are clearly less transitive than others. It is typical of
transitive argument marking that at least one of the arguments is marked differently
from the unmarked S. Explicit argument marking is further a rather obvious
indicator of transitivity, since the explicitly marked argument can usually appear
only in transitive clauses. Consequently, the form of either argument signals
transitivity directly.
If we base our typology only on the number and marking of arguments, we can
distinguish maximally four different types, cf.

1. Neither of the arguments is explicitly marked

2 & 3: Either Agent or Patient is marked
4. Both Agent and Patient are marked

The number of possible types increases significantly, if we take account of the cross-
reference and the possible morphological marking of transitivity in the verb. In the
previous section, we distinguished altogether 24 different transitive clause patterns
that will be represented schematically below. In all the types, the intransitive pattern
is illustrated first followed by the basic transitive pattern. We also use the traditional
labels S, A and O, but somewhat differently from the usual ‘Dixonian sense’. In
addition to the sole argument of intransitive clauses, the label S also refers to the
structurally identical argument in transitive clauses. This should emphasize the
structural similarity of these two arguments (i.e. S is used in a strictly formal sense).
Functionally, the argument in question may be an Agent or a Patient. For example,
the verb agreement in transitive clauses with either of these arguments is referred
to as ‘Agent-‘ or ‘Patient agreement’. The added argument is referred to by A (in
case the Agent is marked) or O (if the Patient is the distinctively marked argument).
V naturally stands for the verb. The letters in subscript refer to cross-reference. The
possible numbers in subscript are used to distinguish between two morphologically
172

identical arguments that behave differently as regards the verb agreement. If V is not
followed by any symbol neither of the arguments is cross-referenced on the verb.
Every schema is followed by a list of features that distinguish transitive clauses from
intransitive ones. Also a brief list of languages that illustrate the type at issue is
given.46 It must also be noted that different subsystems of languages may be encoded
differently and hence exemplify different types. Typical examples include the
different encoding of nominal and pronominal arguments that is attested, for
example, in many Australian languages.47

Quantitative types

Type 1

1. S V
S S V
Transitivity features: Number of arguments
Languages: Kammu, Teribe, Yoruba, Korya Chiini, Thai, Mandarin, Swedish, Nivkh

2. S VS
S S V SS
Transitivity features: Number of arguments and quantitative verb agreement
Language: Alambak

3. S VS
S S2 VS
Transitivity features: Number of arguments
Languages: Iatmul, Ejagham, Spanish, Kilivila, Northern Sotho, Lango, Tubinamba

4. S V
S (X) S V
Transitivity features: Number of arguments, additional (tense) marker
Language: Canela-Krahô

46
The languages listed are not necessarily exemplified in the study.

47
For example, Peter Austin (p.c.) mentioned that 1SG in Jiwarli belongs to type 1, whereas animate
NP:s exemplify type 22.
 173

5. S V AG/PAT
S S V AG+PAT
Transitivity features: Number of arguments and quantitative verb agreement
Languages: Lakhota, Dakota

6. S AG/PAT V
S AG S PAT V
Transitivity features: Number and nature of arguments
Languages: Haida, Central Pomo

Qualitative types

Type 2

7a. S V
S S2 V S2
Transitivity features: Number of arguments, Patient verb agreement
Language: Barai

7b. S V
S S2 VS
Transitivity features: Number of arguments, Agent verb agreement
Language: Päri

8. S VS
S S2 V S2
Transitivity features: Number of arguments, qualitative verb agreement
Language: Chamorro

9. S VS
S S2 V S+S2
Transitivity features: Number of arguments, Patient verb agreement
Languages: Swahili, Kambera, Yimas, Jacaltec, Abaza

10. S V (INTR)
S S V TRANS
Transitivity features: Number of arguments, transitive verb morphology
Language: Bahasa Indonesia
174

11a. S V
S S2 V TRANS S2
Transitivity features; Number of arguments, transitive verb morphology, Patient
agreement
Languages: Toqabaqita, Hoava, Fijian

11b. S V
S S2 V TRANS A
Transitivity features: Number of arguments, transitive verb morphology, Agent
agreement
Languages: Halkomelem Salish, Thompson River Salish

Type 3

12a. S V
S O V
Transitivity features: Number of arguments, marking of Patient
Languages: Ngarluma, Japanese, Korean, Khumi

12b. S V
A S V
Transitivity features: Number of arguments, marking of Agent
Languages: Yalarnnga, Dyirbal, Yidiñ, Lezgian, Pitjantjatjara, Jiwarli, Tongan, Ika,
Samoan

13a. S VS
S O VS
Transitivity features: Number of arguments, marking of Patient
Languages: Finnish, Turkish, German, Russian, Berber, Gamo, Latin, Tamil, Turkana

13b. S VS
A S VS
Transitivity features: Number of arguments, marking of Agent
Languages: Archi, Avar, Hunzib, Balochi

13c. S CL S V
S S2 CL S+S2 V
 175

Transitivity features: Number of arguments, Patient marking on the clitic

Language: W ari

14. S V
A S VA
Transitivity features: Number of arguments, marking of Agent, Agent verb agreement
Language: Kham

15. S V S(INTR)
S O V S(TRANS)
Transitivity features: Number of arguments, marking of Patient, transitive nominative
verb agreement
Language: Kolyma Yukaghir

16. S VS
A S VA
Transitivity features: number of arguments, marking of Agent, Agent verb agreement
Languages: Dolakh ~ New ~r, Nepali

17a. S VS
A S V SA
Transitivity features: Number of arguments, marking of Agent, transitive verb agreement
Languages: Inuktitut, W est Greenlandic, Ngalakan, Yup’ik, Chukchi

17b. S VS
S O V SO
Transitivity features: Number of arguments, marking of Patient, transitive verb
agreement
Languages: Mordvinian, Hungarian

18a. S VS
A S V S+A
Transitivity features: Number of arguments, marking of Agent, Agent verb agreement
(in addition to Patient agreement)
Languages: Mam, W ambaya

18b. S VS
S O V S+O
Transitivity features: Number of arguments, marking of Patient, Patient verb agreement
176

(in addition to Agent agreement)

Language: Tigrinya

19. S1 V S1
A S2 V A(=S1)+O(=S2)
Transitivity features: Number of arguments, marking of Agent, Agent verb agreement
(taken from S), Patient verb agreement (introduced)
Languages: Biri, Sinaugoro, Tukang Besi, W arlpiri, Hula, Lai, Dani

20. S V
A S V TRANS S
Transitivity features: number of arguments, marking of Agent, transitive verb
morphology, Patient agreement
Language: Roviana

21. S1 V INTR S
A S2 V TRANS A(=S1)+O(=S2)
Transitivity features, Number of arguments, marking of Agent, transitive verb structure,
Agent agreement (taken from S), Patient agreement (introduced)
Language: Ngan’gityemerri

Type 4

22. S V
A O V
Transitivity features: Number of arguments, marking of Agent and Patient
Languages: Pitta-Pitta, W angkumarra, Jiwarrli (certain NP’s)

23. S VS
A O VA
Transitivity features: Number of arguments, marking of Agent and Patient, Agent
agreement
Language: Nez Perce

24. S VS
A O V
Transitivity features: Number of arguments, marking of Agent and Patient, loss of verb
agreement
 177

Language: Sindhi

The types illustrated in the previous section can be subdivided into 24 different
types on the basis of the differences between intransitive and transitive clauses. Also
the differences between different intransitive clauses contribute to the division. For
example, types 22 and 24 are identical to each other as regards the transitive clauses,
but are ‘derived’ from different intransitive structures. Furthermore, some cases that
are identical to each other as regards the genuine transitivity features have also been
taken account of. For example, types 1 and 3 mark transitivity identically, but
structurally they differ from each other due to the verb agreement. It is, however,
important to take account of these kinds of difference as well, since it aids us in
distinguishing relevant features from irrelevant ones. Put more concretely, this
means, for example, that we are able to distinguish between languages like Iatmul
and Päri. In Iatmul, the verb agreement is not an indicator of transitivity, since it is
exactly the same in both clause types, whereas in Päri the same feature indicates
transitivity, because it is attested in transitive clauses only.
As the illustration above shows, some types are typologically much more
typical than others. In the following, we discuss some factors that can perhaps
explain the uneven distribution of the types and can perhaps shed some light on why
exactly the kinds of clause type illustrated above are attested in the languages of the
world, while others are not.
The leading principle of the present study is that transitivity is primarily a
semantic notion that is expressed structurally in different ways in different
languages. At the level of morphosyntax, the expression of transitivity is twofold:
the verbs denote events and the arguments refer to participants partaking in them.
The verbs are typically responsible for the semantic nature of events, whereas
arguments refer to the participants in them. The semantic roles of participants
usually inhere in the semantics of verbs and it is usually not necessary to explicate
the roles more closely. The implicit ‘marking’ of transitivity in the verbs also aids
us in explaining the low number of basic clause types in which the verb morphology
(apart from person marking on the verb) is affected by transitivity. The verb is
marked on the basis of (basic) transitivity only in five of the total of 31 different
178

types (including subtypes). This kind of marking seems to be most typical of

Oceanic and Salish languages, since all the languages (except for Ngan’gityemerri)
belong to either of the language groups. Also in these cases, we may refer to the
underlying economy principle of language (see, e.g. Kibrik 1985: 271 among many
others). There is simply no need to mark the verbs by an additional marker that
would just emphasize facets known to us (which is not to say that these kinds of
language are not attested). Marking like ‘he runs (INTR)’ vs. ‘he hits (TRANS) the
boy’ does not reveal anything about the nature of the events that could not be
inferred from the semantics of verbs. If transitivity was a primarily structural feature
of clauses, we would expect there to be more languages in which transitive verbs are
morphologically distinct from intransitive ones. Instead, we find intransitive (e.g.
antipassive, anticausative, passive) vs. transitive (e.g. causative, applicative)
marking of verbs of derived clauses. In derived transitives or intransitives, the
number of arguments and/or participants is somehow different from that inhering
in the basic semantics of verbs and the explicit marking is therefore necessary to
avoid misinterpretations. Furthermore, any kind of purely transitive (i.e. ‘non-
person’) marker on a verb cannot distinguish agents from patients, whereas
argument marking on verbs can. Hence, the mere transitive morphology is not
capable of serving a function that is very typical of transitivity marking.
One of the most significant functions of argument marking is to distinguish
between agent and patient in transitive events. Since transitive clauses involve two
arguments, the language must have some mechanism that distinguishes the two
arguments explicitly from each other. Intransitive clauses, on the other hand, only
have one argument and there is no need to disambiguate the clause. Typically, both
in basic intransitive and basic transitive clauses we can infer the semantic role(s) of
participants from the semantics of verbs. Hence, there is no need to mark the
arguments according to the roles of their referents. For example, we know that
‘breaking’ involves an agent and a patient, even if the arguments are not marked
accordingly. A more plausible explanation for the appearance of explicit argument
marking lies in the discriminating function noted above. Since intransitive clauses
only have one argument, there is no need to separate this argument from anything
else and consequently, no marking appears. On the other hand, even if know that
 179

transitive events involve an agent and a patient we still have to distinguish between
the referents of the roles. Therefore, at least one argument has to be marked. As a
result, explicit marking is restricted to transitive clauses and the mechanism at issue
can be labelled as a transitivity marker.
The predominance of types that can somehow be explained by referring to the
need to disambiguate clauses as regards their semantic role assignment is
overwhelming. Of the total of 31 types, all except five (1-4, 10) can be explained by
referring to this principle. The alert reader may be led to think that we have just
demolished our own principle by claiming that genuine marking of transitivity is
irrelevant, since the marking follows from the need to disambiguate clauses. This
function is equally important in semantically highly transitive (‘killing’, ‘hitting’
etc.) as in less transitive (‘seeing’, ‘loving’ etc.) events, and consequently the
motivation is not found in semantic transitivity. This picture follows from the fact
that the typology illustrated above is based solely on differences between
intransitive and basic transitive events. The goal of the previous section was to show
how transitive events differ from intransitive ones and the approach adopted was the
most appropriate one. We have not taken account of syntactically transitive clauses
that are morphologically less transitive than the basic clauses. That the mechanisms
illustrated above are truly transitivity markers becomes obvious, if we compare the
basic transitive clauses to clauses that profile less transitive events. One example
suffices to clarify this point, cf.

(345) mies kaatu-i

man.NOM fall.down-3SG.PAST
‘The man fell down’

(346) jaakko kaato-i torni-n

Jacob.NOM fall(TR).PAST-3SG tower-ACC
‘Jacob made the tower fall down’
(347) mies ajattel-i lapsi-a-an
man.NOM think.about-3SG.PAST child.PL-PART-3POSS
‘The man thought about his children’ (Finnish)

Examples (346) and (347) both describe events that involve two participants.
180

However, only the latter can be considered typically transitive. Example (347)
denotes a less transitive event, which is reflected in the marking of the Patient that
appears in the partitive case. If the transitive marking was motivated solely by the
disambiguating function, clauses (346) and (347) should both be equally transitive.
However, as shown here the semantic transitivity also contributes to the marking of
transitivity. Also in (347), the less transitive nature of the event denoted inheres in
the semantics of the predicate verb, which means that an explicit marking would not
be necessary. The basic NOM-ACC structure is, however, implicitly related to the
expression of highly transitive events only, because of which the transitive marking
is not possible in (347).
Up to now, we have not said much about the distribution or existence of the
kinds of basic clause exemplified above. In general we may say that the typology
illustrated above reflects the economy principle and the typological nature of
languages. We already explained the rareness of morphological transitivity marking
by referring to the uneconomical nature of this kind of marking. It is more
economical to explicitly mark features that are not directly inferable from the
semantics than to mark basic cases. This principle is also reflected in the typology
above. As noted above, the nature of events inheres in the semantics of verbs in
basic cases. Hence, there is no need to mark any argument according to its role in
a given event. Analytic languages do not have any morphological means at their
disposal to mark arguments for case, because of which their basic transitive clauses
have two identically marked arguments instead of one. The grammatical functions
of arguments are usually distinguished solely by the word order. Word order does
not, however, qualify as a genuine transitivity marker here, since arguments
necessarily appear in a certain order. Furthermore, the order is relevant only in
transitive clauses. The lack of morphology does not result in ambiguity, though. The
chief difference between intransitive and transitive events is the number of
participants partaking in the event and also each of the quantitative types expresses
this difference sufficiently. Languages that are morphologically richer than purely
analytic languages have naturally more mechanisms to explicitly mark different
aspects of transitivity. The marking is also in these cases largely conditioned by the
economy principle. As noted previously, the language has to be able to somehow
 181

distinguish the agent of the event from its target. This is usually achieved by
marking either of the arguments explicitly for its role. The role of the unmarked
argument can be inferred by default. Marking of both would violate the economy
principle. Consequently, languages that mark both arguments explicitly are rarer
than languages that only mark one of the arguments. If the primary function of
argument marking was to emphasize transitivity, we would expect languages like
Pitta-Pitta and Nez Perce to be much more frequent than they are. Even if
transitivity contributes to the marking of basic transitivity, it is still secondary to the
economy principle.
5. A TYPOLOGY OF TRANSITIVITY ALTERNATIONS

In the previous chapter, we proposed a typology of basic transitive clauses. The

illustrated typology was based on explicit differences between intransitive and
transitive clauses. Any such feature the appearance of which is strictly restricted to
transitive clauses was regarded as a transitivity marker in a given language. In this
chapter, we proceed to the main topic of the present study, i.e. to the typology of
transitivity alternations. The goal is to illustrate what kinds of semantic, pragmatic
or primarily structural features of clauses can result in what will be labelled as a
transitivity alternation. Transitivity alternations will be discussed from different
perspectives and many different types will be distinguished. The types at issue
include, for example, transitivizing alternations, intransitivizing alternations,
alternations affecting the number of arguments/participants and general vs. specific
alternations. In the present study, the label ‘transitivity alternation’ is used in a wider
sense than is the usual practice. In many other studies, the notion refers to changes
in the argument structure only. Typical examples include passives, antipassives and
causatives. Many previous studies have focussed on some specific alternation type
(see 1.1.). Our goal is to discuss the typology of alternations as exhaustively as
possible. Consequently, individual alternations are discussed only superficially in
some cases, since a detailed presentation of all possible alternations would make the
study unnecessarily long. Different alternations are usually illustrated only in light
of couple of examples.
A couple of words of caution are in order before proceeding to the typology
itself. The following presentation is as exhaustive as possible at the moment and it
comprises every relevant alternation we have been able to find in our data (of some
250-300 languages). We have included examples from radically different languages
from all over the world, which implies that the vast majority of possible alternations
must be illustrated. Furthermore, many of the illustrated cases are attested in
radically different languages spoken in different parts of the world, which implies
that these kinds of case are certainly relevant for a cross-linguistic typology of
transitivity and cannot be ignored in future studies of transitivity in undocumented
languages. However, since there are features that affect the transitivity only in a very
small number of languages, we cannot exclude the possibility that such features will
be discovered in languages not known to us. This problem is especially acute as
 183

regards the structural typology proposed in 5.2.6. It is logically possible that certain
languages employ very ‘exotic’ mechanisms to signal changes in transitivity and we
have to revise the typology later. The absolutely final word on transitivity has been
said, when all the languages in the world have been studied in detail. However,
since nothing beyond what will be illustrated on the following pages has been found,
we believe to have proposed a rather exhaustive typology. Already after studying the
first 100 languages, we only rarely came across something radically new. The
explanatory principles governing the distribution of the alternations also merit a
detailed study of their own. For example, why are certain kinds of structural
alternation more frequent than others, while some logically possible types are not
attested at all? A detailed analysis of this falls outside the scope of the present study,
even though some points are taken account of. We therefore look forward to future
studies that will undertake this task.

5.1. Transitivity alternation defined

As always, it is in order to start by defining the given notion accurately. Hence, we

are going to use the next dozen pages to illustrate what we mean by the notion of
transitivity alternations. A couple of other definitions are illustrated for the purpose
of showing how our definition differs from them. Since transitivity is both a
semantic and a structural phenomenon, both of these aspects have to be taken into
account. We begin by examining some different definitions, after which we proceed
to demonstrating what the notion ‘transitivity alternation’ refers to in the present
study. It is noteworthy that in the following only definitions that explicitly use the
label ‘transitivity alternation’ or ‘(in)transitivization’ in some form are included.
The purpose of this to give an overview of how this very label has been used in the
literature.
Typically, the label transitivity alternation (irrespective of the exact term)
refers to changes in the verb valency that are explicitly marked by manipulating the
verb morphology. Hence, typical transitivity alternations include passives,
antipassives, reflexives (in different functions) and causatives. This kind of
definition is found (among many others, cf. e.g. Dixon & Aikhenvald 2000b: 6ff)
184

in Drossard (1991: 409, cf. below), Haspelmath (1987: 3) and Guerssel (1986: 12,
51):

[...] während sich der prädikativische Aspekt (die TRAN SITIVIERU N G ) darin
manifestiert, dass bilaterale Beziehungen etabliert, d.h. IN ITIAN TEN - BETRO FFENEN -

Relationen durch morphologischen Aufwand am Verb erzeugt werden, und dies

zumeist auf der Basis eines intransitiven, meist patiensorientierten Verbs. (emphasis
in original)

(1) through (8) are examples for morphological transitivity alternations, i.e. the
verbs [...] form transitivity pairs, where both members contain the same root and the
transitivity/intransitivity is marked by grammatical morphemes. (emphasis in original)

The term Transitivization will be used to designate the derivation of a binary function
(i.e. a two-place predicate) from a unary function (i.e. a one-place predicate.
Just as there is a morphological transitivizing prefix in Berber, there exists a de-
transitivizing prefix (ttw) whose effect is the exact opposite of the transitivizing prefix
ss.

The definition outlined by Drossard refers to what is typically labelled as causative,

whereas Haspelmath’s definition refers to ‘anticausatives’ (this label will be used
in the same way in this study). The definition given by Guerssel takes account of
both types. Despite the differences, the definitions above have in common that they
refer only to transitivity alternations that impinge on the valency of verbs. These
kinds of transitivity alternation are perhaps the most typical ones, and it is
consequently natural that they are usually discussed when transitivity alternations
are at issue.
In the present study, the label is used in a broader sense. Every possible
change in the structure of clauses is regarded as a possible transitivity alternation.
The term is used much in accordance with Hopper and Thompson’s approach in
their 1980 paper (even if they do not use the term transitivity alternation). This
includes alternations in which arguments are eliminated without any morphological
marking on the verb as well as changes in the case frames of arguments (cf. Tsunoda
1985:387). Not every two-actant clause is equally transitive (as has been shown
already above), but different case frames relate to different degrees of transitivity
 185

(cf. also Tsunoda 1985: 387 and Drossard 1991: 411ff). These two different aspects
of transitivity are taken account of also by Drossard (1991: 425f), cf.

T RAN SITIVIERU N G
Variante 1 Input: intransitive Verben bzw. monovalente Verben
Output: transitive Verben bzw. bivalente Verben (in ERG-Sprachen ERG-ABS-
Schemata)
Variante 2 Input: intransitive Verben mit zwei Valenzen
Output: transitive Verben, bei denen ein periphere Partizipant zentralisiert wird,
PRO M O VIERU N G (in ERG-Sprachen Transposition von ABS-OBL- in ERG-ABS-
Schemata)

I N TRAN SITIVIERU N G
Variante 1 Input: transitive Verben mit zwei obligatorischen Partizipanten (in ERG-Sprachen
hoch effektives ERG-ABS-Schemata)
Output: intransitive Verben mit gelöschter AG-oder PAT-Stelle

Variante 2 Input: transitive Verben, bei Kasusmarkierung NOM-AKK- bzw. ERG-ABS-

Schemata
Output: intransitive Verben mit distanzierten Partizipanten, z.B. nun NOM-OBL
für NOM-AKK.

Drossard widens the scope of the notion by including cases that do not necessarily
involve changes in valency. This renders it possible to label the latter versions
(Variante 2) as transitivity alternations as well. It also seems irrelevant whether the
alternations are marked on the verb or not, since Drossard includes examples in
which the alternations are signalled solely on the arguments.
The closest equivalent to the definition proposed here (see below) is what
Kibrik (1993: 49, cf. also 1996: 261) has labelled as a propositional derivation, cf.

I hypothesize that every verb of a given language has its original, basic, prototypical
propositional structure (PS) that can be retained or modified in specific verb
occurrences. The notion of propositional structure of the verb includes several
semantic characteristics – in the first place, the valence structure and the verb’s
semantic class (state/process/action/...). The original propositional structure can be
changed in various ways – an argument can be added to it, or eliminated, or the role
186

of an argument can be replaced, or semantic class characteristics can be changed etc.

These changes are called propositional derivations and within the scope of the
present paper essentially coincide with shifts in semantic transitivity as understood by
Hopper and Thompson 1980. The rationale behind the introduction of propositional
structure/derivation terminology is the wish to avoid the discussion and probable
redefinition of such terms as and notions as voice, diathesis, syntactic relations etc.
which would be necessary if they were applied to the material of the present paper.

According to Kibrik, any change in verbs’ semantics can result in a propositional

derivation. Kibrik uses the term in his 1993 (and 1996) paper to refer to cases in
which the form of the transitivity indicator shifts from Ø- to d- or from »- to l-.
Hence, it seems that we are dealing with a propositional derivation only in cases in
which the verb is explicitly marked. What is important for our purposes is that the
derivation does not imply any changes in valency, but it also applies to cases in
which the valency is not affected (the argument marking may change).
As rightly stated by Kibrik, the inherent semantics of every verb is a rather
complex combination of different facets including semantics, valency etc. (Kibrik
refers to the list of transitivity parameters enumerated by Hopper & Thompson).
Many of these aspects can result in what Kibrik labels as a propositional derivation.
This is also shown by Hopper and Thompson, since they illustrate how the
parameters affect the clause structure in different languages. The definition proposed
by Kibrik enables us to regard many different kinds of verbal change as possible
transitivity alternations. Similarly to the definitions cited above, it is important for
Kibrik that the changes in the propositional structure are explicitly marked by
manipulating the verb morphology (i.e. in the case by changing the transitivity
indicator in Athabaskan). We agree with Kibrik (and also Hopper and Thompson,
Tsunoda and Drossard) in that there is much more to transitivity alternations than
the mere changes in valency. We, nevertheless, think that some modifications are
necessary in order that the definition would be capable of describing every possible
transitivity alternation illustrated in the subsequent sections. In the following, we
demonstrate how we define a transitivity alternation.
In brief, the notion of transitivity alternation refers here to every possible
change in the clause structure (including both semantics and morphosyntax) that can
 187

be explained by referring to some aspect of transitivity. In the most typical cases, the
given features are semantic in nature and coincide with structural ones.48 These
kinds of alternation are, whenever possible, used as a kind of prototype to which less
obvious cases are compared. The goal of this is to avoid circularity. Our definition
of transitivity alternations is rather similar to what Kibrik (cf. above) labels as
propositional derivations. Since transitivity alternations can either increase or
decrease the transitivity, at least one of the constructions involved has to be
transitive. Put concretely, this means that we are dealing with a transitivity
alternation only in cases in which the transitivity of basic clauses is somehow
decreased or in which the transitivization produces a typical transitive construction.
Other cases will be ignored. This approach enables us to include less obvious cases
to our typology. This also means that the features involved in typical transitivity
alternations (e.g. agency, affectedness etc.) are regarded as relevant only if the
alternation in question affects transitive clauses. Unlike Hopper and Thompson
(1980: 253), we do not wish to claim that intransitive clauses can rate higher for
transitivity than transitive ones. We do not distinguish intransitive clauses from each
as regards their transitivity. Different features become relevant only if there are two
participants involved. For example, only the latter of the following examples is
regarded as transitivity alternations, even if both cases exemplify the same semantic
change, cf.

(348) tPo naizdra P qitra

we.PAT to.ground fell
‘W e fell to the ground (not our fault)’
(349) a-tPo naizdra P qitra
AGT-we to.ground fell
‘W e fell to the ground (through our own carelessness)’ (Batsbi, Comrie 1978: 366)

(350) ravii davaaii pii gayaa

Ravi.NOM medicine.NOM drink go.PERF
‘Ravi (impulsively) drank up the medicine’

48
The effects of a semantic change can be reflected only indirectly, but for convenience we focus on
the more salient alternations in this section.
188

(351) ravii-ne davaaii pii daalii

Ravi-ERG medicine.NOM drink pour.PERF
‘Ravi (deliberately) drank up the medicine’ (Hindi, Mohanan 1994: 74)

Since the involvement of two participants is a prerequisite of transitivity in the

present study, cases like (348) and (349) are excluded. On the other hand, the
alternation type illustrated in (351) affects the structural transitivity, in addition to
which the transitivity of the event is decreased. Hence, the likes of (351) will be
taken account. In ideal cases, alternations are restricted to transitive clauses (the
English passive is a rather illustrative example of this kind of alternation). This is
not the case in Batsbi or Hindi, since the illustrated alternation can be attested also
in intransitive clauses.
In principle, every alternation that is structurally identical to a genuine
transitivity alternation (i.e. to a semantically motivated alternation) qualifies as a
possible transitivity alternation. The motivation does not need to be semantic, but
there are also primarily structurally determined cases. However, not everything that
seems as a transitivity alternation is automatically regarded as one. An example of
a ‘false transitivity alternation’ is the following, cf.

(352) õura wanguli-la

dog bark-PAST
‘A dog barked’
(353) õura-õu munda bada-la
dog-ERG snake bite-PAST
‘A dog bit a snake’
(354) õaya barri-la
I cry-PAST
‘I cried’
(355) õaya na õu-na bada-la
I he-ACC bite-PAST
‘I bit him’
 189

(356) õaya munda bada-la

I snake bite-PAST
‘I bit a snake’ (Bidjara, Blake 1976: 282, cited from Breen 1973) 49

In Bidjara, highly transitive events are typically expressed employing absolutive-

ergative (nouns) or nominative-accusative (pronouns) patterns as shown above.
Since the marking in (356) is neither of these, it qualifies structurally as an
intransitivizing alternation. However, the event denoted is equally transitive as
those profiled in (353) and (355). The differences in marking are due to the different
reference to the participants in events. Hence, (356) cannot be regarded as a genuine
transitivity alternation. A further problematic case is provided by the complete free
omission of case marking attested, for example in Yingkarta (Dench 1998: 58). In
this (in principle) ergative language, the ergative marking seems to be omitted
without any obvious reason. In order that this could be regarded as a transitivity
alternation, we should be able to show that the marking is at least somehow
associated with transitivity.
As noted above, the starting point of the definition proposed here is the
semantics. This means that we typically motivate the alternations by referring to
changes in semantic transitivity. Typical parameters include agency, affectedness
of the patient, (in)definiteness of arguments etc., but there are also cases in which
the primary motivation may even be a structural feature. As a consequence, every
change in one of these parameters is regarded as a possible transitivity alternation.
However, even if this definition may seem sufficient, there are cases in which we
run into trouble. First, it is very difficult to define semantic transitivity exhaustively.
Transitivity is a vast phenomenon that covers in one way or another very many
facets of language. In some cases, it is very difficult to say whether a given
alternation is a transitivity alternation or not. In order to define transitivity
alternation meaningfully this way, we would have to restrict the notion accurately.
If we were able to do this, it would be a lot easier to define transitivity alternations
exhaustively as well, since every alternation that is motivated by a change in

49
Example (356) is one of our own and it is based on the marking illustrated in (352)-(355),
consequently we take the whole responsibility for the possible ungrammaticality.
190

semantic transitivity would illustrate a transitivity alternation. There are cases in

which we have to use evidence from other structures to justify our claims (cf.
below). In this respect, the primacy of semantics becomes very obvious. In any case
if possible, we try to justify our decisions by referring to changes in the inherent
semantics of events (examples will follow). The semantically based definition aids
us in judging different alternations correctly in most cases and it has to suffice in the
present context. Hence, the most typical transitivity alternation is a change in the
inherent semantics of the verb that is somehow marked linguistically.. Examples are
consequently the following, cf.

(357) §ò § p vr tó §
I shake table
‘I shake the table’
(358) tó § hmpìr
table ANTIC-shake
‘The table is shaking’ (Kammu, Svantesson 1983: 111)

(359) tuààDii tankhaa vadìì

your.F.SG salary.F.SG increase-PST.F.SG
‘Your salary increased’
(360) malaak ne tuààDii tankhaa vadààii
boss ERG your.F.SG salary.F.SG increase-CAUS-PAST.F.SG
‘The boss increased your salary’ (Punjabi, Bhatia 1993: 239)

(361) ctlcg-e keyõ-cn penr c-nen

father-ERG bear-ABS attack-3SG+3SG+AOR
‘Father attacked the bear’
(362) ctlcg-cn penr c-tko-g §e kayõ-etc
father-ABS attack-ANTIP-3SG+AOR bear-DAT
‘Father ran at the bear’ (Chukchi, Kozinsky et al 1988: 652)

(363) man-nen bygg-de hus-et

man-DEF build-PAST house-DEF
‘The man built the house’
(364) man-nen tänk-te på hus-et
man-DEF think.about-PAST PREP house-DEF
‘The man thought about the house’ (Swedish)
 191

Examples (357) and (358) are the most obvious examples of transitivity alternations.
The alternation is marked on the verb, which is an important facet of transitivity
alternations, even if, as noted above, we do not regard as a prerequisite. However,
verb morphology can aid us in deciding which of two possible alternations is
regarded as one. This is especially important in the case of causatives vs.
anticausatives, since the relevant events are usually equally possible in the world we
live in and the verb morphology is the only valid criterion available. Case marking
is not a criterion here, since arguments are usually more marked in transitive
clauses, which does not imply that we are dealing with a transitivity alternation here.
The construction with a marked verb morphology is regarded as an alternation. The
case is converse in (359) and (360). Example (360) exemplifies a genuine
transitivity alternation. Examples (361)-(364) exemplify alternations that impinge
on a minor transitivity feature. The number of participants and arguments is not
affected. These are also obvious transitivity alternations, since the conditioning
factor is semantic transitivity and the resulting construction is structurally less
transitive than the original one.
One important facet related to transitivity alternations is the semantic
markedness of events at issue. If one of the events is clearly more marked, we may
conclude that the clause employed in the expression of the less typical event
illustrates an alternation. This usually correlates with structural markedness. This
means that in case two constructions can be regarded as alternations, the one that
appears in a less typical case frame or whose predicate is clearly marked is regarded
as altered. Put concretely, this means that in the case of highly transitive events, e.g.
the frame ERG-ABS is seen as basic and ABS-DAT as altered, whereas in the case
of less transitive events ERG-ABS can be regarded as altered, since the original
construction is usually less transitive. There are, however, numerous cases in which
neither of these is adequate, since the constructions involved are equally marked (or
rather, unmarked) both semantically and structurally. This means that we cannot
argue justly for the markedness of either of the constructions. An example of this
is illustrated below, cf.

(365) zamira.di get’e xa-na

Zamira.ERG pot break-AOR
192

‘Zamira broke the pot’

(366) get’e xa-na
pot break-AOR
‘The pot broke’ (Lezgian, Haspelmath 1993: 292)

The examples above are both equally unmarked in all relevant respects.
Semantically they illustrate a causative/inchoative pair in which the clause types are
distinguished simply by the Agent omission or introduction. Both events are equally
possible non-linguistically. We know that people can break things and that things
can break spontaneously without any involvement of an agent.50 There are neither
valid semantic nor valid structural arguments for the higher markedness of either
construction. The verb is genuinely labile in this case, which means in the present
context that it can be used both intransitively and transitively without any
morphological changes. The ergative marking of Agent is not related to markedness
in (365), since this illustrates a typical transitive clause a significant part of which
the ergative case marking of Agent is. We can justly speak of true markedness only
in cases in which the ergative marking is lacking, even if it were expected based on
the basic semantics of the denoted event. This is the case in (367), cf.

(367) zamara.di-waj get’e xa-na

Zamira-ADEL pot break-AOR
‘Zamira broke the pot accidentally/involuntarily’ (Lezgian, Haspelmath: ibidem)

Since it is typical of human agents to act volitionally and deliberately, events like
that profiled in (367) can be regarded as less typical and hence more marked than
the likes of (365). The agent in (367) deviates from our conceptualized agent
prototype due to the lower degree of volitionality. Consequently, (367) qualifies as
an alternation and (365) as the corresponding basic clause.
Cases like (357)-(364) are self-explanatory. We do not have to refer to any
other evidence to justify whether we are dealing with a transitivity alternation or not.
The semantic changes attested are obvious and the events are ontologically distinct.

50
Haspelmath (1993b: 103) is probably right in stating that breaking is typically instigated by an
agent, but this is irrelevant here.
 193

Hence, this suffices to show that we are dealing with a transitivity alternation. The
most obvious transitivity feature is the number of participants and/or arguments.51
Consequently, we can safely argue that clauses having one argument are at some
level less transitive than those having two (cf., however, below). This does not have
to be justified any further, since this structural difference reflects the most
fundamental difference between transitive and intransitive events. Possibly
problematic in this respect are cases in which the arguments are simply eliminated
without any further changes in the clause structure, e.g. he eats vs. he eats bread or
the following kinds of ‘passive’, cf.

(368) dí muj ee díki dí kasába

the woman dig the cassava
‘The woman digs the cassava’
(369) dí kasába díki
the cassava dig
‘The cassava is dug’ (Saramaccan, Job 1985: 166, cited from Glock 1972)

Despite the omission of the Patient or the Agent, (369) remains semantically
transitive and involve two participants one of which is semantically an agent,
whereas the other is a patient. If we view transitivity exclusively as a semantic
phenomenon, this and similar kinds of alternation cannot be viewed as transitivity
alternations. However, the structural differences between the exemplified clauses
are so obvious that cases like (369) will be regarded as transitivity alternations as
well. The other obvious feature of transitivity is the marking of arguments. Most
cases in which the argument marking is affected can be regarded as transitivity
alternations. What makes the argument marking so central for our analysis is that
changes in the inherent semantic transitivity of events are typically expressed by
manipulating case frames. Typical accusative or ergative structures are usually
related to marking of high transitivity, whereas less transitive events are denoted by
other kinds of clause. For example, in Finnish, the nominative-partitive frame is

51
The way in which the arguments are referred to (i.e. are they expressed as independent arguments
or only referred to by person affixes) is irrelevant here, but all instances of ‘intransitive referring’ are
considered.
194

used in experiencer constructions and also in the case of uncompleted events. In this
kind of case, we can justify the lower degree of transitivity of the latter constructions
by referring to the lower inherent transitivity of the former construction. The lower
degree of transitivity of experiencer constructions is used as a criterion.
There are also less obvious cases in which we cannot argue for the lower
transitivity of a structure, even if a given clause seems semantically less transitive
than typical ones, cf.

(370) mies tappo-i karhu-n

man.NOM kill.PAST-3SG bear-ACC
‘The man killed a/the bear’
(371) mies e-i tappanut karhu-a
man.NOM NEG-3SG kill.PAST.PERF bear-PART
‘The man didn’t kill a/the bear’
(372) mies viha-si karhu-a
man.NOM hate-3SG.PAST bear-PART
‘The man hated the bear’ (Finnish)

(373) der mann töte-te den bär-en

art.NOM man kill-3SG.PAST ART.ACC bear-ACC
‘The man killed the bear’
(374) der mann töte-te den bär-en nicht
ART.NOM man kill-3SG.PAST ART.ACC bear-ACC not
‘The man didn’t kill the bear’ (German)

Examples above illustrate affirmative vs. negated transitive clauses of Finnish and
German. As can be seen, in Finnish the negation is an obvious transitivity
alternation. The Patient loses its accusative marking and it appears in the partitive
in negated (originally) transitive (NOM-ACC) clauses. That we are dealing with a
genuine transitivity alternation can be seen if we compare (371) to (372). Example
(372) illustrates a semantically motivated transitivity alternation and since the
negated transitive clauses are structurally similar to this construction, we are entitled
to label (371) as a transitivity alternation. As for the semantics of transitivity,
negated clauses naturally profile less transitive events than affirmative ones, since
nothing has happened and there is neither affected patient nor volitional agent
 195

present. But, what about the examples from German? On the basis of data from
Finnish, we may say that negation is a semantic transitivity feature, since it clearly
affects transitivity in Finnish (Finnish is not exceptional in this respect, but ‘less
transitive negation’ is found also in other languages, including Russian (see e.g.
Payne (J.R.) 1985:231), French, Estonian, Latvian, Paamese and Yukaghir).
However, it seems inappropriate to claim that (374) would be less transitive than
(373), only because it is ‘semantically less transitive’. Consequently, we should
restrict the use of the term transitivity alternation to cases in which changes in the
semantics somehow affect the transitivity of clauses. The difference between
Finnish and German illustrates the obvious and well-known fact that languages
differ crucially from each other in whether a certain kind of change in semantics
results in a transitivity alternation or not. Consequently, we should not try to find
transitivity alternations where they do not exist.
A further example of another kind of possibly problematic case is given
below, cf.

(375) hasan bir öküz ald x

Hasan.NOM a(n) ox bought
‘Hasan bought an ox (non-specific)’
(376) hasan öküz-ü ald x
Hasan.NOM ox-ACC bought
‘Hasan bought the ox’ (Turkish, Comrie 1989: 132)

(377) veo un árbol

see.PRES.1SG ART tree
‘I see a tree’
(378) veo a mi profesor
see.PRES.1SG PREP my teacher
‘I see my teacher’
(379) voy a la escuela
go.PRES.1SG PREP ART school
‘I go to school’ (Spanish)

In both Turkish and Spanish, definite Patients are marked differently from indefinite
ones. In Turkish, it is rather obvious that we are dealing with a transitivity
196

alternation in these cases, since definite Patients appear in the accusative case that
is the typical case of Patients. Consequently, the appearance of this case marker is
restricted to transitive clauses, which makes it to a genuine transitivity marker. In
Spanish, Patients referring to human participants are marked with the preposition
a that is also used to express location, as illustrated in (379). Since definite Patients
appear cross-linguistically in more transitive case frames than indefinite ones, it
seems consequent to regard cases like (378) as more transitive. However, the
mechanism used does not stand in any direct relation to transitivity. Patients in
locative cases are rather related to decreases im transitivity. The relevant question
is, then, can we regard cases like (378) as transitivizing alternations? In the view
that every such change in the clause structure that can be said to be motivated by
some relevant aspect of transitivity, examples like (378) should be included in our
definition. What makes this possibly problematic is that the mechanism used points
rather to intransitivization instead of transitivization. There is a clear disagreement
of function and the corresponding linguistic mechanism, which may make it difficult
to argue justly for the higher transitivity of either clause.
In cases discussed in (368)-(379), the difference between transitive and less
transitive constructions is expressed by manipulating the case marking of the
Patient. The differences are quite obvious (even if possibly problematic, as in (377)-
(379)), since non-basic case frames are usually related to lower degrees of
transitivity. Any deviation from the basic scheme can usually (cf., however, (352)-
(356)) be labelled as a transitivity alternation. Changes in the verb morphology are
usually more problematic in this respect. Also here, there are clear cases. For
example, morphologically marked anticausatives (see, e.g. (358) above) and
agentless passives are less transitive than the corresponding basic clauses. This is
due to the fact that the change results in changes in valency or case frame. But, how
about the following kinds of example, cf.

(380) mies poltt-i talo-n

man.NOM burn.TRANS-3SG.PAST house-ACC
‘The man burned the house’
(381) mies tul-i poltta-nee-ksi talo-n
man.NOM come-3SG.PAST burn.TR-PAST-TRANSL house-ACC
 197

‘The man burned the house accidentally’ (Finnish)

Example (380) represents a typical transitive construction. Example (381) illustrates

a less transitive version of (380), since the degree of agency is reduced. The agent
does not intend the event to happen, but accidentally partakes in it. This difference
is here expressed only by manipulating the verb morphology. The argument marking
remains unaffected. Consequently, if we claim that the case marking reflects the
basic transitivity of events, (379) cannot be labelled as a transitivity alternation.
However, typological evidence shows that the kind of change illustrated in (381)
results in a transitivity alternation in other languages including Guugu Yimidhirr,
cf.

(382) ngayu galga nhanu dumbi

1SG.NOM spear.ABS 2SG.GEN.ABS break.PAST
‘I broke your spear (on purpose)’
(383) ngadhun.gal galga nhanu dumbi-idhi
1SG.ADESS spear.ABS 2SG.GEN.ABS break-REFL.PAST
‘I broke your spear (by accident)’ (Guugu Yimidhirr, Haviland 1979: 125)

Can we use this kind of data to support our claim? The semantic differences
between the two versions of the same basic event are quite obvious, as illustrated
above. Does this suffice to label (381) as a transitivity alternation? What makes
constructions like (381) problematic is that we cannot justify our claims by referring
to the lower transitivity of similar constructions. It is possible to proceed in one of
two ways. First, we can define (381) as a transitivity alternation. This analysis is
supported by the lower semantic transitivity of (381), which is also reflected in the
morphology of the verb. We may ignore changes in the case frame as a precondition,
since semantic changes are not necessarily reflected this way, cf.

(384) mies näk-i talo-n

man.NOM see-3SG.PAST house-ACC
‘The man saw the house’
(385) tuli tuho-si talo-n
fire.NOM destroy-3SG.PAST house-ACC
198

‘The fire destroyed the house’ (Finnish)

Events profiled above are clearly less transitive than typical transitive ones.
Example (384) illustrates an experiencer construction, whereas (385) is
(semantically) more similar to (381), since there is no volitional agent involved.
There are no differences in the case marking of arguments. Cases like (381) are
more obvious and salient transitivity alternations than (384) and (385), since
changes in transitivity are marked by manipulating the morphology of the predicate.
The other possibility is naturally that we ignore the likes of (381). The primary
motivation for this analysis lies in the transitive argument marking. We cannot
justify the less transitive nature of (381) by referring to changes in the case frame.
Furthermore, labelling (381) as a transitivity alternation makes it very difficult to
distinguish it from other structures in which the verb is marked, but which is not
motivated by any facet of transitivity. This includes the following, cf.

(386) hän kaato-i puu-n

s/he fall.down.PAST-3SG tree-ACC
‘S/he fell down the tree’
(387) hän kaataa rysä-ytt-i puu-n
s/he fall.down.INF crash-CAUS-3SG.PAST tree-ACC
‘S/he crashed down the tree’ (i.e. s/he fell down the tree and it made a crashing sound)
(Finnish)

Only (386) exemplifies a typical transitive construction. However, the argument

marking in (387) is identical to (386). Example (387) is similar to (381), since in
both cases the argument marking remains the same despite changes in the verb
morphology. Furthermore, both in (381) and (387) the added component of the
predicate is finite, whereas the ‘original’ verb appears in a non-finite form.
Consequently, as regards the pure structure of these constructions, (387) should also
be regarded as a transitivity alternation, if (381) qualifies as one. This is
semantically quite unfortunate, since in (386) the changes are clearly not
conditioned by transitivity. The additional element rysäytti specifies the nature of
the action described by the predicate. It has nothing to do with any relevant facet of
transitivity and we should consequently be able to disregard it as a transitivity
 199

alternation. Only semantics can aid us in distinguishing between these constructions.

In this study, the first proposed analysis is preferred. This is due to the
primarily semantic approach. Since we define transitive constructions as a
combination of semantics and structure, it seems unjustified to disregard
semantically motivated alternations like (381) in the following typology. On the
other hand, in (387) only the structure speaks for the ‘less transitive’ nature of the
given construction, while the semantic transitivity is unaffected. Hence, (387) will
be disregarded. Even if we regard cases like (381) as transitivity alternations in the
present study, their status is viewed as somewhat lower in comparison with cases
in which the alternation is expressed by manipulating the argument structure. In
cases like (381), we can use the semantics of constructions as the only possible
guideline in our evaluations. Without any direct evidence it might prove difficult to
argue justly for the lower (or higher) degree of transitivity in some cases.
Consequently, we usually have to resort to secondary evidence and use data from
other languages.
Above, we have illustrated our definition of transitivity alternations. It is very
important to bear in mind that a direct comparison of structures is not possible. This
means that we should (if possible) compare two instances of one and the same basic
clause. Two-argument clauses can be semantically less transitive than those
involving only one. We are here not referring to cases like those represented by
Hopper and Thompson (see Hopper and Thompson 1980:253), but we refer to more
fundamental differences. For example, he eats is clearly more transitive than he ran
five miles, even if the structure points to another interpretation. Our definition does
not make it possible to directly compare clauses that ‘originate’ from different
sources with each other. In other words, we can compare clauses he eats meat and
he eats on one hand and he ran and he ran five miles on the other and conclude that
the transitive versions of these are more transitive at some level than the structurally
intransitive ones. These cases differ further from each other in that in the former
case the differences are much more significant than in the latter. It goes without
saying that clauses in which the differential number of arguments is marked by
manipulating the morphology (i.e. valency) of the verb also illustrate transitivity
alternations.
200

Before proceeding to the typology itself, it is in order to say that the definition
proposed above is best regarded as an idealization. Because of the inadequacies in
the data, we will be using some examples that we cannot justify by referring to
semantically motivated examples. This is not to say that the data would be
insufficient, but in some cases the aspects that concern us most are not discussed in
the reference. Despite these problems most examples can justly be used. In most
cases, structures can be regarded as less transitive on the basis of obvious
differences in transitivity. The examples are always compared to either typical
transitive structures (intransitivizing/de-transitivizing alternations) or to typical
intransitive structures (transitivizing alternations).

5.2. Structural/functional aspects of transitivity alternations

In section 5.1., we illustrated what the notion of transitivity alternation refers to in

this study. In the present section, we will dwell on some structural features related
to transitivity alternations in light of cross-linguistic data. The notions to be
discussed include lexical vs. morphosyntactic alternations, general vs. specific
alternations, transitivity increasing, decreasing or rearranging alternations and
alternations affecting the number of participants, arguments and some individual
transitivity features.52 Before proceeding to the individual alternation types, we
discuss the differences between what we have labelled as semantic/functional vs.
structural alternations.

5.2.1. Semantically vs. structurally conditioned alternations

These two alternation types illustrate the most important division related to
transitivity alternations. The former of these refers to cases in which an alternation
is determined by an obvious change in the semantic (typically ontological)
transitivity. Ontologically motivated alternations can be divided into two based on
whether the alternation is conditioned by differences in the inherent semantics of

52
The differences between the last three of these are also largely semantic, but they are included in
the discussion here.
 201

events or by changes in some individual transitivity features.53 The latter type

includes cases in which not only the semantics, but also structural features
contribute to the nature of alternations. A typical example is provided by the uneven
distribution of (morphologically distinct) passives and antipassives. What
distinguishes these from semantically conditioned alternations is that the
contribution of certain structural features is more significant than in semantically
determined cases.
As noted above, the ontology of events makes the most significant
contribution to the expression of transitivity alternations. This means that
ontological differences between events have to be expressed explicitly in order to
avoid obvious misinterpretations (this refers to cases in which semantic information
inhering in the verb does not suffice). As regards the linguistic expression of events,
these kinds of difference necessarily result in linguistic differences, since, for
example, the number of non-linguistic referents changes. Typical examples of
ontologically conditioned cases include therefore e.g. causatives, anticausatives and
reflexives.54 However, not only the ontology of events, but also the semantics
associated with individual verbs contributes to the nature of transitivity alternations.
In ideal cases, there is a direct correspondence between ontology of events and the
semantics of verbs, i.e. an alternation is conditioned directly by the altered semantics
of a predicate (which might be signalled by morphosyntactic changes). In this case,
homophonous verbs behave differently depending on the meaning of the verb. The
form of the verb as such is neutral, since the differences in meaning determine the
alternations. An example is presented below, cf.

(388) y-iwd wryaz taddart

3MS-reach man.CST house
‘The man reached the house’
(389) y-ss-iwd wma aryaz i taddart
3MS-TRANS-reach my brother man to America (sic!)
‘My brother took the man to the house’

53
These are discussed in detail in 5.3.1.

54
These are not discussed in any more detail here because of their obviousness.
202

(390) *t-ttw-iywd taddart

3FS-DETR-reach house
(The house was reached)
(391) y-iwd wryaz adeggwal-enns
3MS-reach man.CST inlaw-his
‘The man wronged his inlaw’
(392) *y-ss-iwd wma aryaz i wdeggwal-enns
3MS-TRANS-reach my brother man to inlaw-his
(My brother made the man wrong his inlaw)
(393) y-ttw-yiwd wdeggwal-enns
3MS-DETR-wrong inlaw-his
‘His inlaw has been wronged’ (Berber, Guerssel 1986: 65f)

As the translations show, the verb (y)iwd can has meanings ‘reach’ and ‘wrong
(TRANS)’. The nature of the transitivity alternation applicable to the clause is
directly conditioned by semantics. If the meaning is ‘reach’, the clause can be
transitivized (i.e. causativized), but not de-transitivized (i.e. passivized). The
situation is converse, if the meaning is ‘wrong’. The verb as such allows both
alternations, but the nature of the alternation varies on the basis of meaning. Hence,
the relevant factor has to be the semantics, not the verb itself.
Differences like those illustrated above are not exceptional, since they reflect
ontological differences. Examples (388)-(393) can consequently be regarded as a
further type of ontologically motivated alternations. They illustrate cases in which
a single verb has two meanings, which results in differences in its syntactic
behaviour. The exact opposite of this is illustrated below, cf.

(394) t-rzem tewwurt

3MS-open door.CST
‘The door opened/the door is open’
(395) t-kkunser tewwurt
3MS-open door
‘The door opened/the door is open’
(396) y(*-ss)-rzem wryaz tawwurt
3MS(*-TRANS)-open man.CST door
‘The man opened the door’
 203

(397) y*(-ss)-kunser wryaz tawwurt

3MS-TRANS-open man.CST door
‘The man opened the door’ (Berber, Guerssel 1986: 48)

(398) anas'i-w t'awla-w-vn s cn ct't'c'c

carpenter-DEF plank-DEF-ACC split.PERF.3M
‘The carpenter split the plank’
(399) t'awla-w tc-s cn ct't'ck'c
plank-DEF ANTIC/PASS-split.PERF.3M
‘The plank split/was split’
(400) anas'i-w t'awla-w-vn fcllct'c
carpenter-DEF plank-DEF-ACC split.PERF.3M
‘The carpenter split the plank’
(401) t'awla-w tc-fcllct'c
plank-DEF PASS-split.PERF.3M
‘The plank was split (*the plank split)’ (Amharic, Amberber 2000: 317)

In these cases, a single event is expressed by two lexemes that differ from each other
in the transitivity alternations they allow. The basic nature of the event remains the
same, which makes these kinds of case more interesting than those illustrated above.
Examples (394) and (395) appear to be synonymous (see Guerssel: ibidem).
However, they behave differently as regards the introduction of arguments. Both
allow causativization, but differ from each other in whether the causativization is
signalled by the verb. The verb rzem is a labile verb that can be used both
transitively and intransitively. Morphological causativization results in
ungrammaticality, as shown in (396). The verb kkunser is the opposite of rzem in
this respect, since the verb has to be causativized in order that an Agent can be
introduced. In (398)-(401) from Amharic the alternation marker is the same, but the
interpretation is sensitive to the verb used. Similarly to (394)-(397) from Berber, the
semantics of particular verbs excludes certain alternations. In Amharic, this is not
marked explicitly, since the marker employed allows in theory both of the
alternations at issue. As opposed to Berber, the differences attested can be explained
by referring to the semantics of the verbs, even if the verbs are very close to each
other in their meaning. The difference here is that fcllct'c implies that an instrument
such as an axe is used, whereas scnct't'c'c is neutral in this respect
204

(Amberber:ibidem). The use of instruments implies an agent, which excludes the

anticausative reading.
In (394)-(401), the very subtle differences in meaning are reflected in the
nature of causativization. The basic clauses are structurally identical. The other
possibility is that the changes are reflected directly in the clause structure. In this
case, the nature of possible transitivity alternations is sensitive to these
morphosyntactic differences, cf.

(402) kenraali lö-i vihollis-en

general.NOM hit.PAST-3SG enemy-ACC
‘The general beat the enemy (in a battle)’
(403) kenraali lö-i vihollis-ta
general.NOM hit.PAST-3SG enemy-PART
‘The general hit an/the enemy’
(404) vihollinen lyö-tiin
enemy.NOM hit-PASS.PAST
‘The enemy was beaten’
(405) vihollis-ta lyö-tiin
enemy-PART hit-PASS.PAST
‘The enemy was hit’ (Finnish)

(406) der mann bau-te ein-en tisch

ART.NOM man build-3SG.PAST ART-ACC table
‘The man built a table (finishing it)’
(407) der mann bau-te an ein-em tisch
ART.NOM man build-3SG.PAST PREP ART-DAT table
‘The man was building a table (without finishing it)’
(408) ein tisch wurd-e gebaut
ART.NOM table become.PAST-1/3SG build.PARTIC
‘A table was built’
(409) an ein-em tisch wurd-e gebaut
PREP ART-DAT table become.PAST-1/3SG build.PARTIC
‘Someone was building a table (without finishing it)’ (German)

In Finnish, the verb lyödä (‘hit/beat’) has two meanings. If the meaning is ‘beat
someone in a battle or a game’, the case frame is the highly transitive NOM-ACC,
 205

as in (402). On the other hand, if the intended meaning is ‘hit’ the clause appears in
a less transitive NOM-PART frame. Examples (406) and (407) illustrate similar
differences. In (406), the event is completed, whereas (407) illustrates an on-going
activity in the past that did not result in a finished house. In both languages these
differences are reflected in the nature of the passive alternation. In Finnish, only
accusatively marked Patients (excluding pronouns) can be promoted to the
nominative in the passive. Consequently, only in (404) is the passivization
‘complete’. Differently from Finnish, German has two different kinds of passive that
are usually labelled as personal (persönlich) and impersonal (unpersönlich) passives
(see e.g. Brinker 1971). Personal passives can be derived only from typical transitive
clauses, as in (408), while all other clause types only allow an impersonal
passivization (if any). In personal passives, the Patient of a transitive clause is
promoted to the subject status. Since the Patient does not appear in the accusative
in (409), it cannot be promoted and the passive is impersonal. In both Finnish and
German, the differences illustrated above are primarily due to differences at the
level of basic clauses, which, on the other hand, follows from semantic differences.
These different passive constructions are (naturally) not restricted to cases like
(402)-(409), but they are attested also in cases in which the basic semantics of the
verb is responsible for similar differences. In this respect these cases are different
from (389)-(393) in which the semantics associated with polysemous verbs (and not
the verb as such) are clearly responsible for which transitivity alternations are
applicable.
In (389)-(409), the nature of the alternations is conditioned either by the
ontology of events, the semantics of verbs or a combination of semantics and
structure, as in the examples from Finnish and German. Moreover, there are cases
in which the nature of an alternation varies depending on the overall nature of an
event (including the ontology of events and the nature of participants involved).
This means that a single alternation is interpreted differently according to the nature
of the original event. A preview owas given in (398)-(401) from Amharic in which
subtle differences resulted in the exclusion of an alternation (semantically). A
further, rather typical example of an alternation whose function depends on the
overall nature of events is provided by the structural polysemy of
206

reflexive/anticausative that is attested, for example, in Finnish, German and

Amharic. The following examples are from Finnish, cf.

(410) jaakko piilo-utu-i

Jacob.NOM hide-REFL/ANTIC-3SG.PAST
‘Jacob hid (himself)’
(411) ovi ava-utu-i
door.NOM open-REFL/ANTIC-3SG.PAST
‘The door opened’

In (400) and (401) from Amharic, the differences in the reading are solely due to
subtle differences in the semantics of the verb. In (410) and (411), on the other hand,
not only the semantics of the verb, but the nature of participants involved also
makes a contribution. Hiding oneself is an activity that is necessarily associated with
a rather high degree of agency. Consequently, the anticausative reading is excluded
in (410). On the other hand, ‘opening’ is an event that is inherent only to doors,
gates etc, i.e. only inanimate entities can open. Both alternations exemplified above
can also be used figuratively or metaphorically (e.g. avaimet ovat piiloutuneet
johonkin ‘the keys have hidden themselves somewhere’ in which an inanimate entity
is interpreted as being capable of acting volitionally, is possible, yet rather
marginal). The semantics of hiding requires that the sole participant be an agent,
while typical undergoers of opening are inanimate entities.
A further semantic facet associated with transitivity alternations noteworthy
in the present context is provided by the consequences of the different motivations
of transitivity briefly discussed in 3.2.2. The differences in the underlying
motivation of transitivity have direct consequences for the status of participants in
events. In inherently transitive events, both participants must have a referent (that
may, however, refer to one non-linguistic participant, as in reflexives).
Consequently, anticausative alternations are excluded. As regards the nature of the
alternations this means, for example, that only a passive interpretation is possible
in the case of inherently transitive events, even if the language in question expresses
passive and anticausative by the same alternation marker. This is the case, for
example, in the following examples from Amharic, cf.
 207

(412) lcm cd-u (b c-aster) tc-k'rr ct'c

rope-DEF (by-Aster) PASS-cut.PERF.3M
‘The rope was cut (by Aster)’
(413) b cr-u tc-k cffctc
door-DEF ANTIC/PASS-open.PERF.3M
‘The door (was) opened’ (Amharic, Amberber 2000: 314)

Examples (412) and (413) are very similar to (410) and (411), but with the crucial
difference that above the conditioning factor is the motivation of transitivity. In
(412), the event denoted is inherently transitive and only a passive reading is
felicitous. On the other hand, in (413), both passive and anticausative readings are
possible due to the nature of the event denoted.
The differences between extended intransitives and inherently transitive
events are also very obvious in this respect. Since both the agent and the patient are
integral parts of inherently transitive events, it is possible to omit any explicit
reference to either participant without this affecting the basic semantics of events
in any way. It is irrelevant how significant the changes may be, since the participants
inhere in the semantics of the verb. On the other hand, the patients of extended
intransitives are only ‘temporarily’ participants in transitive events because of which
they have to be explicitly referred to for the purpose of enabling a transitive reading.
This has been explicitly noted, for example, by Halliday (1967: 47) who states that
clauses like he marched cannot be interpreted as involving an implicit patient, but
in order for extended intransitives to be interpreted as transitive, the patient has to
be explicitly stated (cf. also Keenan 1987:191). In English, the omission of the
patient in extended intransitives coincides with a change in the reading. The other
possibility is that the omission of reference results in ungrammaticality. This is the
case, for example, in Paamese in that the likes of ‘he pushed’ are deemed
ungrammatical (see Crowley 1987: 65).
A further important aspect in which the three different kinds of transitive
event diverge is the degree of agency associated with the events. Inherently
transitive events (at least the prototypical cases) imply a rather high degree of
agency and accidental instigation of events is not possible. These differences also
affect the nature of alternations at least in Finnish and Malay, cf.
208

(414) tul-i poltettua talo

come-3SG.PAST burn.TR.PARTIC house.NOM
‘(I) burnt the house accidentally/(I) effortlessly burnt the house/(I) accidentally burnt the
house (instead of the barn)’
(415) tul-i maalattua talo
come-3SG.PAST paint.PARTIC house.NOM
‘(I) effortessly burnt the house/I accidentally painted the house (instead of the barn)/*(I)
painted the house accidentally’
(416) miehe-ltä puto-si kuppi
man-ABL drop-3SG.PAST cup
‘The man accidentally dropped the cup’
(417) *miehe-ltä maala-utu-i talo
man-ABL paint-REFL/ANTIC-3SG.PAST house.NOM
(The man accidentally painted the house) (Finnish)

(418) dia pecah-kan pasu itu

3SG break-CAUS vase the/that
‘S/he broke the vase’
(419) dia ter-pecah pasu itu
3SG ter-break vase the/that
‘S/he accidentally broke the vase’
(420) dia cat rumah itu
3SG paint house the/that
‘S/he painted the house’
(421) ?die ter-cat rumah itu
3SG ter-paint house the/that
(S/he painted the house accidentally)
(422) dia ter-cat rumah itu lain
3SG ter-paint house the/that different
‘S/he accidentally painted someone else’s house’ (Malay, examples (and the comments below)
courtesy of Foong Ha Yap)

In Finnish, there are three different constructions that can express accidentality of
events. Two are illustrated above. They can express three different meanings all of
which are possible, if the event in question is causative (as in (414)). The completely
accidental reading is excluded, if the event in question is inherently transitive (as in
(415)). The accidentality is here restricted to the choice of the target. The action
itself is purposeful, but the target is not the intended one. Differences in the
 209

expression of decreased agency are also reflected in the ungrammaticality of (417).

The construction illustrated in (416) and (417) only has the meaning ‘do something
accidentally’, and consequently (417) is ungrammatical. In Malay, accidentality is
generally expressed by the verb prefix ter- (that also has other functions). As regards
the consequences of the motivation of transitivity of events for the expression of
accidentality, Malay is very similar to Finnish. The prefix can be used in (418)
without any further modifications. Constructions with ter- become less acceptable,
if the event in question is inherently transitive. In this case, we have to emphasize
the fact that the accidentality is related only to the target, as is the case in (422) that
is interpreted identically to (415).
A further example of the different behaviour of inherently transitive and
causative events are given in the following examples from Tukang Besi (Donohue
1998: 88, 92) and Inuktitut Eskimo (Nowak 1996: 174f, the terms employed in the
discussion of Inuktitut are taken from Nowak if not indicated otherwise), cf.

(423) no-‘ita-‘e na ‘ obu te kalambe

3R-see-3OBJ NOM dog CORE girl
‘The girl saw the dog’
(424) no-‘ita te ‘obu na kalambe
3R-see CORE dog NOM girl
‘The girl saw the dog’
(425) no-buti-‘e na kaluku te anabou
3R-fall-3OBJ NOM coconut CORE boy
‘The boy dropped the coconut’
(426) *no-buti te kaluku na anabou
3R-fall CORE coconut NOM boy
(The boy dropped the coconut)

(427) jaani-up niqi niri-jaa

John-ERG meat.ABS eat.PARTIC.3SG/3SG
‘John eats the meat’
(428) jaani nirijuq (niqi-mik)
John.ABS eat-PARTIC.3SG (meat-COM )
‘John eats (meat)’
(429) arnaq-up titirauti-Ø siqumit-tanga
woman-ERG pencil-ABS break-3SG/3SF.TR.NOM.PARTIC
210

‘The woman broke the pencil’

(430) *titiraut siqumit-tuq arna-mit
pencil.ABS break.PARTIC.3SG woman-ABL
(W oman broke a pencil)

Both in Tukang Besi and Inuktitut, the inherently transitive predicates allow more
alternations than causativized ones. In Tukang Besi (inherently) transitive verbs
allow both the subject and the object to appear in the nominative, as shown in (423)
and (424). This is restricted to these kinds of verbs, since many verbs that can be
used both transitively and intransitively only allow nominative objects. Hence, (426)
is ungrammatical. The situation is very similar in Inuktitut, since the ‘absolutive
shift’55 is restricted to activity predicates and it is not possible in the case of result
predicates. Result predicates, as ‘break’ and ‘spill’ do not allow oblique arguments
in intransitive forms, as shown in (430).56
Above, we have illustrated some examples of semantically conditioned
alternations. In some of these, as in (402)-(409) the structure of clauses also makes
a significant contribution to the nature of applicable alternations. Also the basic
structure of languages contributes to the morphosyntactic nature of individual
alternations. Perhaps the most obvious example of this is provided by the
distribution of morphologically marked (i.e. ‘typical’, cf. e.g. Siewierska 1984: 2,
Givón 1989: 113) passives and antipassives in the languages of the world. The
former is more typical of predominantly nominative-accusative languages, whereas
the latter is more frequently attested in absolutive-ergative languages. Both of these
functions can naturally be expressed in both language types, but the nature of the
alternation is determined by the basic structure of a given language. The rationale
behind the dominance of either alternation is best explained by referring to
differences in the status of arguments in transitive clauses. In predominantly
accusative languages, the Agent is the primary argument in transitive clauses. The
grammatical status of the Patient is lower, which is reflected, for example, in the

55
As noted by Nowak (1996:174) the term comes originally from W oodbury 1975.

56
In light of the examples illustrated the division into activity and result predicates seems to overlap
with what we have labelled as inherently transitive and causative verbs.
 211

fact that the Patient can be freely eliminated (e.g. he eats bread vs. he eats). In
ergative languages the situation is converse, i.e. the zero marked Patient is typically
the primary argument. Hence, the Agent is freely omissible. In both language types,
the omission of the primary argument is usually accompanied by other changes.
Typically, the secondary argument is promoted to the primary status, which is
marked by manipulating the morphology of the verb. In case the original primary
argument is expressed, its status has clearly decreased and it usually appears in an
oblique form (e.g. he was killed by the man). The promotion of the secondary
argument may follow from the demotion of the primary one, which has been
regarded as the primary function of passive, for example, by Shibatani (1985: 837).
This is the case at least in languages with a strong subject relation.57 In case the
primary argument is demoted some other argument has to take its place. Since the
subject in transitive clauses of accusative languages is the Agent, the omission of
Agent results in the promotion of the Patient (which is usually accompanied by
morphological changes in the verb). The situation is the exact opposite of this in
ergative languages. What is important for our purposes is that the basic argument
marking pattern of languages has direct consequences for the structural nature of
alternations. The function or the alternation is the same (i.e. the demotion or
promotion of Agent/Patient), but the consequences for the structure of clauses are
clearly different depending on the dominant argument marking pattern of languages.
Above, we briefly discussed the generally known fact that accusativity of a
language usually implies the existence of a morphological passive, whereas
ergativity is typically related to the presence of a morphological antipassive.58 These
differences in the marking of arguments are also reflected in the expression of
semantically conditioned alternations. As has been noted above, accusative and
ergative languages differ from each other in the transitivity aspect they emphasize
linguistically. In accusative systems, the linguistically marked facet is the
affectedness parameter, whereas ergativity focusses on the agency. Hence, we might

57
‘Subject’ is here used in a purely structural sense, i.e. it refers to the (usually zero marked) sole
argument in intransitive clauses and to the argument of transitive clauses that is marked in the same
way.

58
This is a broad generalization with numerous exceptions.
212

expect that accusative systems are more sensitive to marking changes in the
affectedness, while ergative languages stress the importance of agency for high
transitivity.59 For the structure of transitivity alternations this has the consequence
that it is more ‘convenient’ for accusative languages to express changes in the
affectedness parameter, whereas ergative languages more easily express changes in
the agency. In other words, in accusative languages changes in the affectedness are
iconic as are changes in the agency in ergative languages. A couple of rather
illustrative examples are given below60, cf.

(431) už-~ ’ikay y-exu-r-si

boy-ERG glass.ABS II-break-CAUS-PAST.W IT
‘The boy broke the glass’
(432) uži-q ’ikay y-exu-s
boy-POSS glass.ABS II-break-PAST.W IT
‘The boy accidentally broke the glass’ (Tsez, Comrie 2000: 365)

(433) ich hab-e den teller zerbrochen

I.NOM have.PRES-1SG ART.ACC plate break.PARTIC
‘I broke the plate’
(434) mir ist der teller zerbrochen
I.DAT be.PRES.3SG ART.NOM plate break.PARTIC
‘I broke the plate accidentally’ (German)

(435) mielipuoli tappo-i nuo ihmise-t

maniac.NOM kill.PAST-3SG those people-NOM(=ACC)
‘The maniac killed the/those people’
(436) mielipuoli tappo-i ihmisi-ä
maniac.NOM kill.PAST-3SG people-PART
‘The maniac killed (some) people’ (Finnish)

(437) hansi-p inuit tuqup-paa

Hansi-ERG people kill-IND.3SG.3SG
‘Hansi killed the people’

59
A detailed analysis of this merits an indepedent study.

60
It is in order to note that the following examples are typical cases and there are naturally exceptions.
 213

(438) hansi inun-nik tuqut-si-vuq

Hansi.ABS people-MOD kill-ANTIP-IND.3SG
‘Hansi killed people’ (W est Greenlandic, Manning 1996: 82)

Examples (431)-(434) illustrate alternations due to agency. The alternation

expressed is the same, but the alternations in question display clear structural
differences. In the predominantly ergative Tsez, the exemplified alternation only
affects the marking of Agent that shifts from ergative to possessive, whereas the
Patient remains in the zero marked absolutive case (changes in the verb morphology
are ignored, since they are not relevant for the discussion here). In German, this
change is expressed also by manipulating the marking of Agent. However,
differently from Tsez, the marking of Patient is also affected. This follows from the
Agent demotion. Since the Agent is demoted to the dative, another argument has to
be promoted to the subject. A mere change in the Agent is deemed ungrammatical
(i.e. clauses like *mir ist/hat den Teller zerbrochen are not possible). As shown in
(435)-(438), on the other hand, accusative languages express changes related to the
affectedness (or Patient) more easily (i.e. with fewer structural changes) than
ergative languages. Also here, the motivation is the same (i.e. the definiteness vs.
indefiniteness of Patient). In Finnish, the only difference is found in the marking of
the Patient that is in the accusative, if the Patient is definite, while the partitive is
related to indefiniteness. In West Greenlandic, the same alternation produces more
radical changes, as shown in (438). Similarly to (434) from German, the primary
argument has been morphosyntactically demoted and another argument takes its
place. Hence, the differences attested are also best explained by referring to the
nature of the primary argument. In ergative languages the primary, i.e. the zero
marked argument, is the Patient, which results in ‘iconic’ agency-related changes.
On the other hand, in accusative languages the primary argument is the Agent and
changes in agency usually result in changes in the Patient as well.

5.2.2. Lexical vs. morphosyntactic alternations

The second distinction made here is between lexical and morphosyntactic

alternations. The labels are rather self-explanatory. Lexical alternations are
214

alternations in which the relevant changes in transitivity are primarily expressed by

changing the verb lexeme61 that is most directly responsible for the expression of
events. Lexical alternations are often accompanied by morphosyntactic alternations
as well (see below). They are therefore combinations of lexical and morphosyntactic
features. What amounts to labelling them as (merely) lexical alternations is the fact
that the lexemes employed may be said to be primarily responsible for the marking
of transitivity (see below). Morphosyntactic alternations, for their part, are
alternations that are expressed solely by manipulating the structure of clauses. Both
of these types are illustrated below, cf.

(439) absö yön-a

greens cook(3SG)-DECL
‘The greens have cooked’
(440) n absö wc-n-a
1SG greens cook-FUT(1SG)-DECL
‘I will cook the greens’ (Haruai, Comrie 1993: 319)

(441) c'q'int-d kot-¥ilk'e-Ø da wir-s

boy-ERG PRV:3OBJ-push-3:SG:OBJ sister-DAT
‘The boy pushed his sister’
(442) že p-Ø sgalo-x-emtkwe-Ø 'q'int'-s
dog-NOM PRV-3:OBJ-get used-3:SG:AOR boy-DAT
‘The dog got accustomed to the boy’ (Svan, Testelec 1998: 42, cited from Klimov & Alekseev
1986: 13ff)

(443) ngaa-§i waa § saay§-naa-ng

1SG-ERG bird.ABS hear-PRES-1SG
‘I hear a bird’ (non-agentive)
(444) ngaa-§i waa §-kaay § saay§-naa-ng
1SG-ERG bird-DAT hear-PRES-1SG
‘I listen to a bird’ (Chepang, DeLancey 1981: 634)

(406) der mann bau-te ein-en tisch

ART.NOM man build-3SG.PAST ART-ACC table

61
Some discussed cases may also comprise suppletive changes. W hether we are dealing with distinct
lexemes is not obvious in all cases.
 215

‘The man built a table (finishing it)’

(407) der mann bau-te an ein-em tisch
ART.NOM man build-3SG.PAST PREP ART-DAT table
‘The man was building a table (without finishing it)’
(445) der mann dachte an ein-en tisch
ART.NOM man think.PAST.3SG PREP ART-ACC table
‘The man thought about the table’ (German, (406) and (407) repeated here for convenience)

(446) mawun-du ga `i `ilmu man-i

man-ERG tree break-PAST
‘A man broke a tree’
(447) ga `i `ilmu ira
tree break-PAST
‘A tree broke’ (Djaru, Tsunoda 1987: 192)

The examples from Haruai illustrate a typical inchoative/causative pair in which the
differences between the given events are expressed primarily by changing the
lexemes. The lexeme yön can refer only to intransitive event of cooking, whereas
wc implies that the intransitive event is caused by an external agent. Since the event
profiled in (440) is transitive, the number of argument is affected as well. Examples
from Svan illustrate cases in which two events are marked differently on the basis
of semantic transitivity. Only (441) denotes a highly transitive event, whereas the
semantic transitivity of (442) is considerably lower, which results in a decrease in
structural transitivity. In these cases the morphosyntactic changes reflect semantic
differences. Examples from Chepang, German and Djaru illustrate typical
morphosyntactic alternations in the spirit the term is employed here. Since there are
no lexical changes, morphosyntax is alone responsible for expressing the changes
at issue. In Chepang, the alternation is related to changes in agency, whereas in
German the relevant parameter is affectedness (and/or the degree of completion).
In (446) and (447), the morphosyntactic changes are more radical, since the number
of arguments is affected. The only difference between (446)-(447) and (439)-(440)
is that in the former case the lexemes employed are different. Examples from
German and Djaru show that the distinction between the alternations is not clear-cut,
since they can be expressed using the same mechanisms.
An alert reader may at this point criticize the label chosen to describe the
216

lexical alternations. As can be seen above, lexical alternations also involve

morphosyntactic changes. Consequently, a better term for these kinds of alternations
may be lexical-morphosyntactic alternations. We adopted the label for convenience,
since it enables us to make the distinction made above. Another possible source for
criticism is found in the following kinds of example, cf.

(448) nainen tappo-i koira-n

woman.NOM kill.PAST-3SG dog-ACC
‘The woman killed the dog’
(449) nainen näk-i koira-n
woman.NOM see-3SG.PAST dog-ACC
‘The woman saw the dog’ (Finnish)

The events at issue differ crucially from each other as regards their semantic
transitivity. Only (448) can be regarded as typically transitive, while (449) profiles
an experiencer-stimulus event (or relation). The only difference between (448) and
(449) lies in the verb lexeme responsible for the expression of the given semantic
differences. Since no morphosyntactic differences are attested despite the clear
semantic differences, alternations like these could be regarded as ‘genuine lexical
alternations’. These differ from ‘lexical-morphosyntactic’ ones due to their purely
lexical nature. The criticism may further be considered justified in light of the fact
that cases like (449) will be labelled as indirect alternations in 5.3.2. It suffices to
say here that in these cases the decreased transitivity manifests itself indirectly. In
Finnish this means that (449) can be passivized only if the stimulus is considered
somehow affected by the event denoted. In case there are no indirect changes, either,
the likes of (448)-(449) cannot be regarded as any kind of transitivity alternation.
Above, we stated that we classify every possible change in the semantics and/or
structure of clauses as a possible transitivity alternation. Consequently, we should
classify (449) as one, too. However, in the case of pure lexical alternations we are
not dealing with mere changes in transitivity, but far more important here are the
general semantic differences between events. The chief function of the different
lexemes is to distinguish the events from each other, not on the basis of their
transitivity, but due to their general nature. In the case of lexical alternations (in the
 217

sense the term is employed above), we can more justly claim that one of the
functions of different lexemes is also to signal differences in transitivity, since there
are other changes in the clause structure as well. In the latter case, some of the
relevant transitivity features are lexicalized parts of lexemes, which brings about the
kinds of change illustrated. In a way, the lexeme and the morphosyntactic marking
‘share the responsibility’ for the expression of transitivity. However, we do not wish
to restrict the use of the term ‘lexical alternations’ only to these cases, but in our
terminology the term covers both ‘pure lexical alternations’ and ‘lexical-
morphosyntactic’ ones.
Lexical and morphosyntactic alternations differ from each other in how the
overall marking of transitivity is distributed. In morphosyntactic alternations, the
employed lexeme is primarily responsible for the expression of basic transitivity,
while the morphosyntactic marking expresses typically some individual aspects that
are somehow different from the basic cases. In the case of lexical alternations, on
the other hand, many significant aspects are inherently expressed by the lexemes
themselves and changes in the morphosyntax usually emphasize these. There are
also cases in which the morphosyntactic marking of some lexicalized features is
deemed ungrammatical or the interpretation may vary from the intended one.
Examples of this are given below, cf.

(193) koira kuol-i

dog.NOM die-3SG.PAST
‘The dog died’
(448) nainen tappo-i koira-n
woman.NOM kill.PAST-3SG dog-ACC
‘The woman killed the dog’
(450) mies tapa-tt-i karhu-n
man.NOM kill-CAUS-3SG.PAST bear-ACC
‘The man made someone kill the bear’ (Finnish) (ungrammatical for ‘the man killed the bear’)
(Finnish, (193) and (448) repeated here for convenience)

(451) y-ggwu wghrum

3MS-knead bread.CST
‘The bread is kneaded’
218

(452) y-ggwu wryaz aghrum

3M S-knead man.CST bread
‘The man kneaded the bread’
(453) *y-ss-ggw wryaz aghrum
3M S-TRANS-knead man.CST bread
(The man kneaded the bread) (Berber, Guerssel 1986: 27)

These examples demonstrate that it is not possible to express integral parts of

lexemes morphosyntactically. The causativization is an integral part of the lexeme
tappaa (‘kill’) because of which (448) has two explicitly expressed arguments.
Example (450) shows that it is not possible to causativize an inherently causative
lexeme, if the intended meaning is the one illustrated in (448). Example (450) does
not have a monocausative reading. The examples from Berber are somewhat
different. In Berber, the lexeme ggwu is genuinely labile. Since both meanings
inhere in the semantics of the verb, causativization is ungrammatical, as shown in
(453).
The presentation above leads us to another aspect related to lexical and
morphosyntactic alternations. As illustrated and discussed above, certain facets of
transitivity are lexicalized parts of lexemes. As for transitivity alternations, this
means that in the case of lexical alternations the lexeme is primarily responsible for
the expression of transitivity and the possible morphosyntactic changes simply
reflect these features. This also has the consequence that the morphosyntactic
changes resulting from lexical alternations are obligatory, if the given language uses
morphosyntactic mechanisms to express these kinds of difference. This kind of
marking corresponds to what was above labelled as ‘explicit marking’. The
morphosyntactic marking is secondary to the lexical one. It does not have any major
function of its own, since the relevant aspects are computable from the verb.
Morphosyntactic alternations are very different in this respect. Also in this case, the
lexeme plays an important role in the interpretation of clauses, since it is primarily
responsible for the description of events. However, the morphosyntactic changes do
not merely emphasize some inherent aspects of events, but the explicit marking is
related to deviations from the basic transitivity of events. Either the aspect is not an
integral part of the lexeme employed, and morphosyntactic marking is needed to
 219

make a certain reading possible (e.g. morphological causativization) or the event

described differs semantically from the basic one and the morphosyntactic marking
is needed to ensure the intended reading. Because of the nature of changes
expressed, morphosyntactic alternations are usually related to minor changes,
whereas lexical alternations can express a variety of different alternations. Lexical
alternations are typically related to the expression of transitivity of different events,
whereas morphosyntactic alternations usually express changes between different
instances of the same basic events. This also limits the number of alternations
expressed (purely) morphosyntactically.
The generalization outlined above holds true in many cases, but there are also
obvious and rather frequently attested counter-examples. First of all, different
lexemes do not have to differ from each other in any way as regards the transitivity.
Furthermore, not only the number of affected parameters is important, but
morphological alternations occur in cases in which the meanings expressed cannot
be derived from each other. Morphological causatives and anticausatives are good
examples of cases in which the number of parameters affected is rather high, but in
which the different meanings are mutually inferable from each other. Consequently,
there is no need to employ different lexemes. A couple of somewhat different cases
are illustrated below, cf.62

(454) nga ]l-læ nuknuk e

I wash-CMPL clothes the
‘I finished washing the clothes’
(455) nga owo nuknuk læ
I wash clothes CMPL
‘I finished washing clothes’ (Kusaiean, Hopper and Thompson 1980: 259, cited from Sugita
1973: 399)

(456) žek’-~ biš wa r-ac’-xo

man-ERG food.ABS IV-eat-PRES
‘The man eats the food’

62
This pattern is quite common cross-linguistically, especially for verbs of consumption and digestion
and verbs of washing (Peter Austin, p.c.). Also in ‘lexically impoverished’ European languages cases
are attested, e.g. speisen vs. essen in German and dine vs. eat in English.
220

(457) ¨ali Ø-iš-xo

Ali.ABS I-eat-PRES
‘Ali eats’ (Tsez, Comrie 2000: 363, 366)

(458) nau ku fanga naqa

1SG 1SG.NFUT eat(INTR) PERF
‘I’ve finished eating’
(459) nau ku qani-a botho...
1SG 1SG.NFUT eat-3SG.OBJ pork
‘I ate pork...’ (Toqabaqita, examples courtesy of Frank Lichtenberk)

Examples (454)-(459) illustrate lexical alternations in which the lexemes used are
semantically very closely related, but are clearly distinct lexical verbs. In a way,
(454)-(459) exemplify cases in which two lexemes are used to describe subtly
different instances of an event type. However, what makes it justified to label these
examples as instances of lexical alternations is that not only are the lexemes
different, but there are also morphosyntactic changes involved. Consequently,
structurally alternations exemplified above are typical instances of lexical
alternations. What makes them special is the semantic closeness of the lexemes
involved. On the other hand, we might even claim that (454)-(459) illustrate the
only kind of genuine lexical alternations, since the differences in marking follow
primarily from the verb, and not from different degrees of semantic transitivity.
To sum up. In the present section, we have examined lexical vs.
morphosyntactic alternations. Of these two different transitivity alternation types,
lexical alternations are usually related to the expression of transitivity of different
events. In many languages only highly transitive events are expressed employing
typical transitive constructions. Less transitive ones are used in the expression of
other events. Since different case frames are related to different events (denoted by
different verbs), we are entitled to say that the verbs are primarily responsible for
the marking of transitivity in these cases. The possible morphosyntactic changes
only reflect the transitivity of verbs. Morphosyntactic alternations, on the other
hand, express minor differences of the same basic events. The event itself remains
the same, but there are minor changes. We can predict that the expression of
transitivity alternations is distributed this way. Lexical alternations express
 221

alternations and semantic differences that need to be expressed by different lexemes.

We cannot infer very many different meanings from one morphosyntactically
manipulated lexeme or clause, but we simply have to resort to other mechanisms.
On the other hand, morphosyntactic alternations impinge on only minor transitivity
features that are a part of the ‘transitive potential’ of verbs, but which has to be
expressed explicitly, since the reading is somewhat different from the typical one.
Using different lexemes in these cases is uneconomical and is hence more rarely
attested, but is not unheard of.

5.2.3. General vs. specific transitivity alternations

In the previous section, we distinguished between lexical and morphosyntactic

alternations. The distinction was based on whether the lexemes used or some
morphosyntactic mechanism is primarily responsible for the alternations. The
semantics of these alternations is further distributed differently. As a generalization,
we may say that the number of parameters affected as a result of a lexical alternation
is higher, whereas morphosyntactic alternations typically impinge upon individual
features. On the other hand, lexical alternations are by definition restricted to very
specific event types, while morphosyntactic ones are more general. This brief
summary leads us to the topic of this section. Morphosyntactic alternations can
further be divided into general and specific alternations. Lexical alternations are
excluded from the description, since they are related to individual events and cannot
be considered either general or specific according to the view adopted here. The
generality vs. specificity is here based on the number of alternations expressed.
General alternations are expressed by polysemous affixes the function of which
varies according to the environment and function they are used in. Specific
alternations are expressed by monosemous mechanisms with a single function.
Hence, the number of alternations expressed is higher in the case of general
alternations. Both of these alternation types will be illustrated and their nature
discussed in what follows. We begin the illustration by presenting general
alternations, since this is in accordance with the approach adopted in this study.
Before proceeding, we would like to draw the reader’s attention to the fact that
222

many of the alternations labelled as general below are ‘specific’ in the sense that
there may be also other ways to express the alternation in the given language. An
example of this is illustrated by the difference between direct and indirect
causatives. As generally known (cf. e.g. Comrie 1989: 172), languages tend to use
morphological and lexical causatives for direct causation, whereas less direct
causation may be expressed periphrastically. Consequently, in a very strict sense we
cannot label a morphological causative marker as an instance of general alternation
marker, since there is also specific semantics involved (i.e. the directness of the
causation). However, a causative marker that is used regardless of transitivity of
clauses or the animacy of the causee, for example, is semantically more general than
causatives that are distinguished on the basis of these features. In these cases the
generality is motivated indirectly by a comparison to other languages. We will
distinguish five different main types of both alternation types. The division is based
primarily on what kinds of general alternation are attested in languages. In 5.2.3.2.,
specific alternations are discussed in light of this division as well.

5.2.3.1. General alternations

In brief, general transitivity alternations are morphosyntactic alternations that are not
related to a single function, but which can express a variety of transitivity
alternations. The alternations are structurally identical in all the uses (typically the
verb affix or the case marking is the same), but the specific function varies, for
example, depending on the (semantic or structural) environment. The term
‘transitivity alternation’ refers in this section mainly (but not exclusively) to
individual mechanisms (e.g. verb or case affixes) that impinge on the clause
structure. We have adopted a somewhat looser use of the term in order to be able to
illustrate the very general nature of some mechanisms. For example, this enables us
to label verb affixes that can either increase or decrease the transitivity of clauses
as general markers. This is not possible, if we restrict the use of the term to
constructions instead of individual markers.
General alternations can by definition appear in more than one semantic and
syntactic environment, which implies that they are semantic ‘poor’ or unspecified.
 223

A very specific semantics severely restricts the number of semantic environments

in which an alternation is applicable. On the other hand, we may say that the
semantic poorness follows from the occurrence in many different environments. It
is very difficult to definitely state which is the cause and which the result. Luckily
enough, this is not our concern, but it suffices for our purposes that general
alternations are semantically somewhat ‘bleached’ in comparison with specific ones.
General alternation markers acquire their specific semantics from the environment
they appear in. Semantic roles and the nature of participants along with the
semantics of events make a significant contribution to this. General alternations can
be seen as a further manifestation of the economy principle noted numerous times
in this study. It is not necessary or it may even be regarded as redundant to explicitly
mark features that can be inferred without an explicit marking.
The mere introduction or omission of arguments can be regarded as the most
general of all possible transitivity alternations. This comprises numerous different
functions, such as plain transitivization or intransitivization, non-morphological
passives and antipassives, non-morphological applicatives, anticausatives (in a non-
morphological sense) and so on. We do not go to details here, but it has to suffice
to merely mention the type.63 In the discussion, we will focus on alternations that are
somehow marked morphologically as well. These often include introduction or
omission of arguments, but these are always accompanied by morphological
changes in the verb. General alternations are dividable into five types depending
on the semantic explicitness of the kind of (structural and semantic) alternations
expressed. Three of the types are somehow related to changes in valency, whereas
the fourth type is neutral in this respect and the last one expresses only minor
changes. All of these will be examined in what follows, so a more detailed
description of the types is not necessary at this point.
The first type is the least specific and the primary function of the alternations
is merely to mark changes in transitivity without any ‘restrictions’. One and the
same mechanism may relate to both increased and decreased transitivity. A couple
of examples are illustrated below, cf.

63
This is not to say that we would have ignored the mere changes in the number of arguments as a
transitivity alternation type.
224

(460) otoko ga shinda

man NOM die.PAST
‘The man died’
(461) onna wa otoko ni shin-are-ta
woman TOP man DAT die-PASS-PAST
‘The man died on the woman’ (lit. the woman was died by the man)
(462) onna ga otoko o koroshita
woman NOM man ACC kill.PAST
‘The woman killed the man’
(463) otoko wa (onna ni) koros-are-ta
man TOP (woman DAT) kill-PASS-PAST
‘The man was killed (by the woman)’ (Japanese)

(464) k ‚ saram-i na-rul po-ass-ta

DET person-NOM 1SG-ACC see/look-PAST-DECL
‘The man saw me’
(465) k ‚ saram-i po-i-ess-ta
DET man-NOM see/look-PASS-PAST-DECL
‘The man was seen’
(466) k ‚ saram-i na-eykey kurim-ul po-i-ess-ta
DET man-NOM 1SG-DAT picture-ACC see/look-CAUS-PAST-DECL
‘The man showed me the picture’ (Korean, examples courtesy of Soon Mi Hong-Schunka)

(467) mak-a
open-TRANS (Suffixation of the root vowel)
‘To open (TRANS)’
(468) mak-ke
open-ke
‘To open (INTR)’
(469) ray
die
‘To die’
(470) ray-ke
die-TRANS
‘To kill’ (Ainu, Tamura 2000: 211f)

Japanese and Korean are similar to each other in that the same verb affix can either
omit an argument or add one. (460) exemplifies the typical intransitive pattern and
 225

(462) the typical transitive pattern of Japanese. The Japanese examples illustrate so-
called indirect and direct passives (respectively). Structurally, (461) and (463) are
identical to each other. Both constructions have an argument marked as topic
(roughly, ‘subject’) and an argument that is followed by the dative postposition ni
(that in both cases refers to some kind of ‘instigator’ or ‘causer’). Furthermore, in
both cases the verb is morphologically passive. The only structural feature that
distinguishes these two clauses is the predicate. In (461), the verb intransitive
shin(are)ta, whereas in (463) the verb is the lexical causative of this, i.e.
koros(are)ta. These differences are due to different underlying structures, which also
affects the nature of the functions expressed. In (461) a malefactive argument is
added to the clause. The original argument appears in the dative. The situation is
converse in (463), since an argument has been omitted (or rather made optional).64
This argument may be optionally expressed in the dative. The function of the
alternation is not to introduce an argument, but to make a core argument optional.
We are here dealing with the same morphological mechanism (i.e. passive marking
of the verb) that expresses functions contradictory to each other. In the case of an
originally intransitive verb, passivization renders it possible to add an argument to
the clause, whereas the same marker makes an argument freely omissible, if the verb
is transitive. The exact nature of the alternation is not due to the marker or even to
the nature of the participants involved, but the number of arguments is conditioned
by the function expressed by the marker -rare. Examples from Korean are in many
respects very similar to those from Japanese. The affix -i can either passivize or
causativize clauses. In the former case, we are naturally dealing with a transitivity
decrease, whereas the latter exemplifies a transitivity increase. Examples from Ainu
are also ‘transitivity neutral’ and the functions expressed opposite to each other. In
(468), the suffix -ke derives an intransitive verb, whereas in (470) the result is a
transitive verb. In Ainu, the same marker can be said to have both anticausative or
causative functions. In distinction from Japanese and Korean, the differences are
lexically determined.
In Japanese, Korean and Ainu, a single marker may either introduce or omit

64
In Japanese, arguments can be rather freely eliminated also without passivization.
226

arguments. On the basis of our findings, it seems that markers like -rare, -i or -ke
are cross-linguistically rare, even if they are not unheard of (see also Dixon 2000:
31 (Sonrai) and Plungian 1993 (Dogon), cf. also Haspelmath 1990: 46ff). More
typical are general markers related either to increases or decreases in transitivity (but
not both). The alternations may cause changes in valency or they may be more
generally related to transitivity (in this case there are not necessarily any changes in
the verb valency). There are three subtypes of this type (types 2-4 below) all of
which can be subdivided into valency increasing and valency decreasing types. In
the first type (type 2), an alternation is expressed by one and the same marker
associated with either an increase or a decrease in valency. The exact functions vary.
These alternations are illustrated below, cf.

(471) di naa la toog-al-al nenne bi

I AUX you seat-CAUS-APPL child the
‘I will make the child sit down for you’ (W olof, Comrie 1985: 330)

(472) §umu-a-§a
die-3SG-IND
‘He died’
(473) §umu fei-fe-a-§a
die 3SG-TR-3SG-IND
‘He killed him’
(474) wate e §i-i-§a
house make-3PL-IND
‘They built a house’
(475) wate e §i ya-fe-i-§a
house make 1SG-TR-3PL-IND
‘They built a house for me’ (Tauya, MacDonald 1990: 196)

(476) yeni mang-ala kin-anne:

Jenny TRANS-get NOM-eat
‘Jenny is getting rice’
(477) yeni mang-ala-ing kin-anne: adi-nna
Jenny TRANS-get-APPL NOM-eat brother-3.POSS
‘Jenny is getting rice for her brother’
 227

(478) jambu mi-rène

guava INTR-fall
‘The guava fell’
(479) matius mar-rène-ing jambu
Matthew TRANS-fall-APPL guava
‘Matthew dropped guava’ (Seko Padang, Payne 1997: 191, examples courtesy of Tom
Laskowske)

(480) mamai-a bi-soko-ki hida makari

mother-ERG 3SG-wash-NONTHEME DEM clothes
‘Mother is washing clothes’
(481) soko-a-ki hida mamai
wash-DETR-NONTHEME DEM mother
‘Mother is washing’
(482) soko-a-ki hida makari
wash-DETR-NONTHEME DEM clothes
‘The clothes are being washed’ (Paumarí, Chapman & Derbyshire 1991: 298)

(483) t’crmus-u b c-lij-u tc-s cbb cr c

bottle-DEF by-boy-DEF PASS-break.PERF.3M
‘The bottle was broken by the boy’
(484) b cr-u (b c-t’inik’k’ak’e) tc-k cffctc
door-DEF (with-care/attention) ANTIC/PASS-open.PERF.3M
‘The door opened/was opened (with care)’
(485) aster t-at’t’cb c-
Aster REFL-wash.PERF-3F
‘Aster washed herself’
(486) s cww-o -u irsb crs-a cw(*-in) tc-d cbadd cb-u
person-PL-DEF each.other-POSS.3PL(*-ACC) RECIP-hit.RECIP.PERF-3PL
‘The people hit each other’ (Amharic, Amberber 2000: 315ff)

Examples from Wolof, Tauya and Seko Padang illustrate the polysemy of causatives
and applicatives (see Austin 1997 for the discussion of this polysemy in Australian
Aboriginal languages).65 In (471) from Wolof, the affix -al is attached twice to the
same verb in a different function. In (473) from Tauya, the affix -fe- causativizes the

65
Applicative is here used to refer to the general nature of the alternation in question.
228

intransitive (472), whereas (475) exemplifies the use of the same affix in an
applicative function. In Wolof and Tauya, the transitivity of the original clause
seems to be irrelevant. The case is clearly different in (476)-(479) from Seko
Padang. In this language, the function of the marker -ing varies depending on
transitivity. If this affix is attached to a transitive verb, it functions as an applicative
marker, as in (477). On the other hand, with some intransitive verbs the function is
causative, as shown in (479)66. In all these languages, the form remains constant
despite the changes in the specific function. Examples from Paumarí and Amharic
illustrate kinds of ‘mirror image’ of (471)-(479), since these examples illustrate
general transitivity decreasing alternations. The examples above differ from each
other in the number of different alternations expressed by a given marker. In
Paumarí, the same marker can express both antipassive and passive functions. The
only difference between (481) and (482) is the semantic nature of the argument
omitted. In Amharic, the variety of alternations expressed by the affix t/te- is
somewhat greater, since the same marker expresses passive, anticausative, reflexive
and reciprocal functions. This polysemy is not exceptional, but is attested in other
languages as well. Especially, the structural polysemy of passive and anticausative
or passive and reflexive is frequent. This is very understandable, since the functions
are very similar, and it is uneconomical to have several markers for similar functions
that can be inferred from the context.
In the examples above, different instances of valency increase and decrease
are expressed by a single affix. The appearance of this kind of alternation is not
unexpected, since the functions of the alternations are usually contextually inferable.
Type 3 of general alternations is also related to changes in valency. Differently from
(472)-(486), the functions are the same, but there are some finer semantic
distinctions that are ignored. A couple of examples of this are illustrated below
followed by a discussion, cf.

(487) y-n-nuqqeb wattas

3MS-INCH-hole bucket.CST
‘The bucket got pierced’

66
It is not discussed by Payne which verbs take the causative sense.
 229

(488) y-ss-n-nuqqeb wryaz attas

3M S-TRANS-INCH-hole man.CST bucket
‘The man pierced the bucket’
(489) y-bedd wrba
3MS-stand boy.CST
‘The boy stood up’
(490) y-ss-bedd wryaz arba
3MS-CAUS-stand man.CST boy
‘The man made the boy stand up’ (Berber, Guerssel 1986: 17, 14)

(491) gagri puT-ik

vase break-3SG.PAST
‘The vase broke’
(492) sita-« gagri puT-a-ik
Sita-ERG vase break-CAUS-3SG.PAST
‘Sita broke the vase’
(493) b ccca-« macho k ha-ik
child-ERG fish eat-3SG.PAST
‘The child ate the fish’
(494) ama-« b ccca-ke macho k hw-a-ik
mother-ERG child-DAT fish eat-CAUS-3SG.PAST
‘The mother fed the child fish’
(495) aite-« sita-b hi gita-ke ciT hi lek h-a-ik
Aite-ERG Sita-ABL Gita-DAT letter write-CAUS-3SG.PAST
‘Aite made Sita write a letter to Gita’
(496) aite-« sita-b hi gita-ke ciT hi lek h-a-e lcga-ik
Aite-ERG Sita-ABL Gita-DAT letter write-CAUS-INF adhere-3SG.PAST
‘Aite made Sita make someone write a letter to Gita’ (Bote, examples courtesy of Balaram
Prasain)

(497) no-mate-ako te buti

3R-die-APPL CORE fall
‘They died in a fall’
(498) no-hugu-ako te poda-no
3R-chop-APPL CORE knife-3.POSS
‘They chopped with their knifes’
(499) no-ala-ako te ina-su te kau
3R-fetch-APPL CORE mother-1SG.POSS CORE wood
‘She fetched the wood as a favour for my mother’ (Tukang Besi, Donohue 1999: 225ff)
230

(500) siellä tanssi-tt-iin

there dance-PASS-PAST
‘There was dancing’
(501) mies tape-tt-iin (vahingossa)
man.NOM kill-PASS-PAST (by accident)
‘The man was killed (by accident)’
(502) kirja anne-tt-iin häne-lle
book.NOM give-PASS-PAST s/he-ALL
‘The book was given to him/her’ (Finnish)

Examples from Berber and Bote exemplify causatives, those from Tukang Besi
different applicative constructions and (500)-(502) are examples of the Finnish
passive. Berber has only one transitivizing morpheme (ss-) that is used irrespective
of animacy or transitivity (see Guerssel 1986:36). In Bote, the causative morpheme
-a is used irrespective of the transitive valence of the original clause. Berber and
Bote are far from being exceptional languages in this respect, but numerous other
languages (perhaps even the majority of them) have only one causative affix that is
insensitive to transitivity (cf., however, Turkish below). There is perhaps more
variation based on animacy, since causatives in which the causee is an animate
entity are usually somewhat less direct than causatives with non-human patients.
Having only one causative affix is very natural, since the function expressed is in
principle the same, i.e. an agent is added to the original event in every case.
Languages like Wolof, Tauya and Seko Padang are somewhat exceptional, since the
role of the introduced argument is different, even if we are dealing with a valency
increase in both cases. Examples (497)-(499) from Tukang Besi illustrate three
semantically different applicative constructions. The role of the introduced argument
varies. In (497) the added argument is a locative or circumstantial, in (498) it is an
instrument and in (499) a beneficiary. Despite these obvious semantic differences
the applicative affix is the same. This is what Peterson (1998:39) has labelled as
‘non-morphologically distinct applicative construction markers’. The function
shared is the introduction of an argument to the clause core.67 Examples from
Finnish illustrate cases in which the use of the passive morpheme is not conditioned

67
Tukang Besi also has other, morphologically distinct applicatives which will be illustrated later.
 231

by the transitivity nor has the animacy of the Patient or the intentionality of the
(omitted) agent any effect on the structure of passives.
The three types illustrated thus far differ from each other in the status and
nature of generality. We are obviously dealing with general alternations in (460)-
(486), since the functions are different. In (487)-(502), on the other hand, the basic
function is the same and, hence, the generality much more ‘restricted’. Above, we
defined a general alternation as a transitivity alternation whose semantic
unspecificity allows it to express more than one transitivity alternation type.
Consequently, it may seem that we should exclude the latter kinds of case from the
description here. In these cases, we cannot justify the generality of the alternations
illustrated by referring to evidence from the languages in question only, but we have
to resort to typological data. In cases like (472)-(486), mere semantic/functional
evidence suffices, since it is very easy to distinguish between the different
alternations. The semantic and grammatical consequences of the alternations are
clearly different despite certain common properties. For example, no one would
seriously question the semantic differences between anticausatives and reciprocals.
If these are expressed by the same mechanism, we are clearly dealing with an
alternation that qualifies as general in our definition. On the other hand, the
alternations illustrated in (487)-(502) are either semantically or grammatically
identical, because of which mere semantic or grammatical evidence does not suffice.
In order to show that we are dealing with general alternations also here, we have to
be able to present cases in which these (or very similar) kinds of alternation are
explicitly distinguished from each other. The examples that justify this analysis are
given below in 5.2.3.2. where we also illustrate other kinds of specific alternation.
In (460)-(486), the generality is explicit or (‘language-internal’), whereas (487)-
(502) illustrate a more implicit (‘cross-linguistic’) generality.
All the examples so far illustrate general changes in valency. The alternations
differ from each other in whether we are dealing with valency increase or decrease
and also the consequences for the semantics of events are different. The fourth type
of general alternation illustrates a more heterogeneous type. Not only do the
illustrated alternations express changes in valency, but the same markers also
express minor semantic changes, cf.
232

(503) yi-Ø-Ø-béézh
AFX-3.A-TI-boil
‘It is boiling’
(504) yi-Ø-»-béézh
3.G-3.A-TI-boil
‘S/he is boiling it’
(505) yoo-Ø-§©
3.G/3.A.AFX-TI-see
‘S/he sees him/her’
(506) yi-Ø-ní-»-§©
3.G-3.A-AFX-TI-see
‘S/he looks at (examines) him/her’
(507) yi-Ø-i-Ø-nod
3.G-3.A-AFX-TI-lick
‘S/he licks it’
(508) yi-Ø-»-naad
3.G-3.A-TI-lick
‘S/he is licking it’(Navajo, Kibrik 1993: 53, 57f, cited from Young and Morgan 1980:
757,779, 420, 361)

(509) mwana-’yu w-a-dy-a

child-this he-TENSE-eat-INDIC
‘The child has eaten’
(510) mai a-ku-dy-ets-a mwana
woman she-TENSE-eat-CAUS-INDIC child
‘The woman is feeding the child’
(511) mwana-’yu w-a-dy-ets-a
child-this he-TENSE-eat-CAUS-INDIC
‘This child has eaten too much’ (Chichewa, Hopper and Thompson 1980: 264, examples
courtesy of Lee Trithart)

(512) õatu yinana muduwa-yi

1SG.A 2SG.O scratch-PRES
‘I scratch you’
(513) õani muduwa-tadi-yi
1SGS scratch-REFL-PRES
‘I scratch myself’
(514) nulu kana-li nina õanti tayi-yi
3SG.NF.A person-ERG 3SG.NF.O meat.ABS eat-PRES
 233

‘The man is eating this meat’

(515) nawu kana nina õanti tayi-tadi-yi
3SGNFS person.ABS 3SGNFO meat.ABS eat-ANTIP-PRES
‘The man is having a feed of his meat’
(516) õatu yinana danka-na wara-yi
1SG.A 2SG.O find-PARTIC AUX-PRES
‘I found you’
(517) õani danka-tadi-na wara-yi yiõka õu
1SG.S find-ANTIP-PART AUX-PRES 2SG.LOC
‘I found you (accidentally, i.e. I ran into you)’
(518) dala ki·Ûa-tadi-yi õapa tudu-yali
skin peel-PASS-PRES water fire-ERG
‘The skin peeled off because of the boiling water’ (Diyari, Austin 1981: 152ff)

(519) saya mem-buka pintu itu

I ACT/TRANS-open door that
‘I opened that door’
(520) pintu ter-buka oleh angin
door ACCID-open by wind
‘The door was opened by the wind’
(521) maaf, kop i saudara ter-minum oleh saya
sorry, coffee your ACCID-drink by me
‘Sorry, I accidentally drank your coffee’
(522) pintu itu ter-buka
door that STAT/RES-open
‘The door is open’ (Bahasa Indonesia, De Vries 1983: 157, 167, 169)

Examples from Navajo and Chichewa illustrate transitivity increasing alternations,

whereas (512)-(522) exemplify general alternations that decrease valency or affect
some minor facets of transitivity. Both in Navajo and Chichewa, the causative affix
can also be used to intensify a transitive action (Kibrik (ibidem) labels the kind of
transitivization as ‘action intensification’). This is also attested in Kambera and
Godoberi. Furthermore, in Navajo, the same affix is also an agentivizer, which
refers to a shift from experiencer to agent, as in (506). The affix -tadi of Diyari has
valency a decreasing function as shown in (513). In addition, the marker expresses
both antipassive and passive functions ((514) and (515)). In (517), a human agent
acts less volitionally than in typical cases, whereas in (518) the instigator of the
234

event is an inanimate entity incapable of acting volitionally. It is perhaps noteworthy

that in (518) the marking of arguments has not changed. In a closely related Pama-
Nyungan language Yidiñ, the affix -di- has the same functions as -tadi in Diyari (cf.
Dixon 1977: 275). The functions of ter- in Indonesian are similar to those expressed
by tadi-. In addition to the two passives, the affix also expresses resultativity.
The last type comprises cases in which the given alternations are generally
related to minor changes. In distinction from all the previously illustrated types, the
last type has no valency-related functions, but both the original and the resulting
structure are transitive. In theory, all the alternations that are not due to changes in
the inherent transitivity of events, but which can be explained by referring to it,
exemplify this type. Two examples are examined below, cf.

(523) s ~ ‘ai e le ali’i le i’a

PAST eat ERG ART high.chief ART fish
‘The high chief ate the fish’
(524) E alofa le teine i le tama
GENR love ART girl LD ART boy
‘The girl loves the boy’
(525) na va’ai e le fafine le pule= ~’oga i l=o=na
PAST see ERG ART woman ART rule=school LD ART=POSS=3SG
ofisa
office
‘The woman saw the principal at her office’
(526) na va’ai le fafine i le pule= ~’oga i le maketi
PAST see ART woman LD ART rule=school LD ART market
‘The woman saw the principal at the market (accidentally)’ Samoan (Mosel & Hovdhaugen
1992: 423f, 731)

(527) mies rikko-i lasi-n

man.NOM break-3SG.PAST glass-ACC
‘The man broke the glass’
(528) mies rakastaa mehu-a (*mehu-n)
man.NOM love.PRES.3SG juice-PART (*juice-ACC)
‘The man loves (the) juice’
(529) mies jo-i mehu-n
man.NOM drink.PAST-3SG juice-ACC
 235

‘The man drank the juice’

(530) mies jo-i mehu-a
man.NOM drink.PAST-3SG juice-PART
‘The man drank (some of the) juice’ (Finnish)

In Samoan, the absolutive-ergative pattern is associated with typical transitive

events, as can be seen in (523). Also some less transitive events are also expressed
by this pattern, as shown in (525). If a given event displays a clearly lower degree
of transitivity, the pattern shifts to absolutive-locative. The Patient is the explicitly
marked argument in this case. The same pattern is also employed, if the degree of
agency is lower than expected. An example is illustrated in (526) in which the
seeing is not planned, but happens spontaneously. In Finnish the nominative-
accusative pattern is the typically transitive one, while the partitive marking of the
Patient signals lower transitivity. The marking can either be related to a lower
overall transitivity of events (as in (528)) or it can express lower degrees of patient
affectedness, as in (530). In the latter case, the basic transitivity of events is not
affected.
As stated above, we can explain the appearance of general alternations related
to changes in valency by referring to functional similarities. The semantic nature of
events and participants involved also make a contribution. This is also the case in
(523)-(530). In cases like (524) and (528), transitivity as a whole is affected: the
degrees of agency and affectedness are inherently lower than in typical transitive
events. This can be directly inferred from the verb semantics and it is also reflected
in the linguistic expression of events. In (526) and (530), one of the relevant
transitivity features is impinged on. In (526), the degree of agency is lower than in
(525), while (529) and (530) display different degrees of event completion or patient
affectedness. We may say that the generality is here organized on a ‘part-whole
basis’, since the alternations exemplified in (526) and (530) share one feature with
those related to changes in the inherent transitivity. The relation between the nature
of events and the nature of affected transitivity parameters is quite straightforward
in both cases. ‘Seeing’ is an event that does not affect its target in a salient manner,
whereas drinking implies a rather high degree of agency. As a result, the same
alternation can be used in both these cases without any danger of misinterpretations.
236

For example, even if (530) is structurally similar to (528) we know, based on the
semantic nature of the given event, that only the affectedness parameter has been
affected. Hence, there is no need to employ distinct alternations for functions that
can be expressed by a single mechanism.

5.2.3.2. Specific alternations

In the previous section, we distinguished five different kinds of general transitivity

alternation. The generality was understood as a lack of specificity or as an ignorance
of some semantic distinctions relevant in other languages. As is inferable from the
chosen label, specific alternations illustrate the opposite of the general ones. This
means that an alternation is specifically related to the expression of one function
only. This brief definition of the term may be somewhat misleading, since it might
be difficult to find alternations that have only one transitivity-related function
without any further functions whatsoever. Minor overlaps are thus accepted here.
For our purposes, it suffices that given languages has the means to distinguish the
illustrated alternations formally. It is irrelevant whether there are additional ways of
doing this or whether certain specific alternations can in some cases express more
than one function. We will examine the specific alternations by comparing them to
the five general alternation types illustrated above. This means that there are also
five different kinds of specific alternation that are examined in the same order as the
general alternations in 5.2.3.1. The exact functions expressed may differ from those
illustrated above, since we may have not been able to find one-to-one
correspondences between examples from different languages. Prior to the concrete
illustration, we define the given notion in greater detail.
It was stated above that general alternations are a further aspect of the
economy principle of language. The existence of general alternations follows from
the cooperation of semantics and structure. Different alternations that are
distinguishable without any linguistic marking do not need to be distinguished
morphosyntactically. This results is ‘semantic poorness’ of the markers, since the
exact function is inferred largely from the context (including semantic roles of
participants and the nature of events denoted). General markers have the semantic
 237

potentiality of expressing two or more different, yet usually closely related

alternations. The semantics of events excludes the less appropriate ones. A typical
example is illustrated by the reflexive/reciprocal isomoprhism. In this particular
case, the number of the Agent is decisive. Specific alternations illustrate naturally
a mirror image of general alternations in this respect. The semantics of the
alternation is an integral part of the employed marker itself. Thus, the markers can
appear only in favourable environments, i.e. only when the semantics of the marker
and the given alternation agree. As for reflexive vs. reciprocal this means that the
reciprocal alternation is excluded and results in ungrammaticality, if the Agent is
singular. The feature ‘plural agent’ inheres in the semantics of the marker. On the
other hand, if the semantics of a given event allows multiple readings (e.g.
(agentless) passive and anticausative) a change in the marker coincides with a
semantic change. Both of these are illustrated below, cf.

(531) mies pes-i itse-n-sä

man.NOM wash-3SG.PAST self-ACC-3.POSS
‘The man washed himself’
(532) miehe-t pes-i-vät itse-n-sä
man-NOM.PL wash-PAST-3PL self-ACC-3.POSS
‘The men washed themselves’
(533) *mies pes-i toise-n-sa
man.NOM wash-3SG.PAST each.other-ACC-3SG.POSS
(*The man washed each other) (Finnish)

(534) der mann öffnete die tür

ART.NOM man open.PAST.3SG ART door
‘The man opened the door’
(535) die tür öffnete (sich)
ART.NOM door open.PAST.3SG (REFL/MIDDLE)
‘The door opened’
(536) die tür wurde geöffnet
ART.NOM door become.PAST.3SG open.PARTIC
‘The door was opened (by someone)’ (German)

Example (531) illustrates the (transitive) reflexive of Finnish. The reflexive is

238

derived by substituting the Patient with a reflexive pronoun. The reflexive marker
is neutral as regards the number of the Agent. Consequently, (532) is also
grammatical. However, as can be seen in (533) the reciprocal marker requires that
the Agent be plural. Substituting the Patient of a transitive clause with the
accusative reciprocal marker toisensa results in ungrammaticality, which coincides
with the semantics of reciprocals. In (535) and (536), a change in the marker does
not result in ungrammaticality, but in a difference in reading. Differently from
general alternations (cf. (484) and (485) from Amharic) only one reading is possible
in (535) and (536). There is no variation, but (535) can be understood only as
anticausative, while (536) only allows a passive interpretation. Specific alternations
mark explicitly what general alternations express implicitly.
In the previous section, we stated that the kinds of alternation attested in
Japanese, Korean and Ainu are rather rare typologically. This is due to the very
obvious differences in the nature of the alternations. For example, in Ainu the
alternations expressed are the exact opposites of each other. These kinds of
difference cannot usually be inferred from the context, but the differences have to
be explicit. Therefore, it follows that these alternations are expressed by distinct
markers, which is the norm in the languages of the world. It is of the utmost
importance that they are expressed, since they correspond to obvious non-linguistic
differences. Hence, the number of specific alternations in this case is rather high and
the number of general alternations very low. Examples of specific alternations of the
first type are easy to find. An example is given below, cf.

(537) annem kapi-yi aç-tv

mother.my door-ACC open-PAST
‘My mother opened the door’
(538) kapi aç-vl-di
door.NOM open-ANTIC-PAST
‘The door opened (Turkish, Haspelmath 1987: 2)
(539) hasan öl-dü
Hasan.NOM die-PAST
‘Hasan died’
(540) ali hasan-i öl-dür-dü
Ali Hasan-ACC die-CAUS-PAST
 239

‘Ali killed Hasan’ (Turkish, Comrie 1989: 175f)

Turkish is a language with distinct causative and anticausative constructions. This

is cross-linguistically rather frequent and the great majority of languages distinguish
between these two obviously different alternation types.
The second type of specific alternations comprises cases in which the
alternations are closely related, but yet somehow functionally different. The
differences between general and specific alternations are very obvious also in this
case. General alternations are possible, since the general function (an increase or
a decrease in transitivity) is the same despite other differences (e.g. the semantic
roles of introduced/eliminated arguments are different). The context or the
semantics of events usually solves the possible ambiguity. There are differences
despite the similarities, and it should not come as a surprise that languages express
these alternations distinctively. A couple of examples are provided below, cf.

(541) no-wila na anabou i jambata

3R-go NOM father OBL jetty
‘The father went to the jetty’
(542) no-pa-wila te anabou i jambata na ama
3R-CAUS-go CORE child OBL jetty NOM father
‘The father sent the child to the jetty’
(543) no-wila-ako te ina-no i daoa
3R-go-APPL CORE mother-3.POSS OBL market
‘She went to the market for her mother’ (Tukang Besi, Donohue 1999: 212, 232)

(544) buklat ne
open 3SG.ERG/3SG.NOM
He’ll open it’
(545) mam-uklat ya
ANTIP-open 3SG.NOM
He’ll open up (as a shop or house)’
(546) mibu-buklat ya
MIDDLE-open 3SG.NOM
‘It is opening (as door by itself)’
(547) maka-buklat ya
PASS-open 3SG.NOM
240

‘It has been opened/it is open’ (Kapampangan, Mithun, 1994: 257f)

Examples from Tukang Besi illustrate two different transitivizing alternations,

namely causative and applicative that are both expressed by a ‘semantically
specified’ verb affix. The reading varies depending on the marker employed.
Causative and applicative are both valency-increasing alternations, but they diverge
as regards the nature of the introduced argument. Consequently, there is nothing odd
in the fact that these two alternations are morphologically distinct in many
languages. Examples (544)-(547) illustrate explicit marking of antipassive, middle
voice (anticausative in our terminology) and passive in Kapampangan. In all these
cases, the transitivity of the basic event decreases. Kapampangan expresses
explicitly what is only implicit in Amharic.
The appearance of specific alternations illustrated in (537)-(547) is self-
evident as a result of the obvious semantic and functional differences. Even if many
languages are more ‘loyal’ to the economy principle, there is nothing extraordinary
in the specificity of these alternations. Far less natural and necessary are specific
alternations equivalent to type three of general alternations. In these cases, the
functions are the same, but there are, however, languages that use specific
alternations in these kinds of case. We have subdivided type 3 into three subtypes
depending on the semantic nature of these additional features. The first subtype
comprises cases in which the transitivity (transitive valence) of clauses conditions
the use of alternation markers. The markers themselves do not express any major
differences in semantic transitivity, cf.

(548) man-nen måla-r huse-t

man-DEF paint-PRES house-DEF
‘The man paints the house’
(549) man-nen dansa-r
man-DEF dance-PRES
‘The man is dancing’
(550) huse-t måla-s
house-DEF paint-PASS
‘The house gets painted’
 241

(551) huse-t bli-r målat

house-DEF become-PRES paint.PARTIC
‘The house gets painted’
(552) där dansa-s
there dance-PASS
‘There is dancing/people dance there’
(553) *där bli-r dansad
there become-PRES dance.PARTIC
(There is dancing) (Swedish)

(554) Hasan öl-dü

Hasan.NOM die-PAST
‘Hasan died’
(555) Ali Hasan-i öl-dür-dü
Ali Hasan-ACC die-CAUS-PAST
‘Ali killed Hasan’
(556) müdür mektub-u imzala-d x
director letter-ACC sign-PAST
‘The director signed the letter’
(557) diÕçi mektub-u müdür-e imzala-t-tx
dentist letter-ACC director-DAT sign-CAUS-PAST
‘The dentist got the director to sign the letter’ (Turkish, Comrie 1989: 175f)

(558) mu-may-bore burun-bak-yeñ jugu-mana-bore-yi§

MU-food-3NSG 3SG/3NSG-OP-put PAST FEM-mother-3NSG-ERG
‘Their mother set down their food for them’
(559) õun-bata-me mu-wapawapa § mu-õondo-yi§
3SG/1SG-OP-get MU-clothes MU-wind-ERG
‘The wind took (picked up) the dress from me’
(560) Ø-bak-gaw §-miñ jugu-gowko-nowi-ka §
3SG/3SG-OP-call-PAST FEM-grandmother-his-LOC
‘He sang out to his grandmother’ (Ngalakan, Merlan 1983: 48f)

Examples (548)-(553) illustrate two different passives of Swedish. Swedish also has
a third passive (vara (be)-passive) that is more stative in nature. It is ignored here,
since it differs from bli and -s passives also in other relevant transitivity aspects. Bli
and s-passives can be both be considered processual in their meaning. There are no
major differences in meaning between (550) and (551). What is important for our
242

purposes is that only s-passive can passivize intransitive clauses.68 Since the
function expressed by these two different passives (i.e. the agent demotion) is in
principle the same, we cannot explain the differences exemplified in (552) and (553)
by referring to any individual transitivity feature69, but we have to state that their
use is sensitive to (perhaps among other things) the transitivity of the original
clause. Examples from Turkish illustrate two causative affixes: -dür causativizes
intransitive and -t transitive (as well as ditransitive, see Comrie:ibidem) clauses. The
basic function is also here the same, since both causatives introduce an agent to a
given event. There are, however, certain differences, since in (555) a ‘principal
agent’ (i.e. the participant that carries out the action) is introduced, while in (557)
the added agent is a causer that is not necessarily involved in the main event of
signing the letter. The last three examples illustrate two different object promoting
affixes of Ngalakan. Both -bak- and -bata- can be used to express ‘that there is an
animate (usually human) notional ‘indirect object’ cross-referenced in the verb by
object pronominals’ (Merlan: ibidem). In addition, -bak- can be used to enable the
introduction of an additional argument. As regards the transitivity, these markers are
distributed much in accordance with bli and -s passives of Swedish, since the
markers have common functions.
In the second subtype of type 3, the use of different markers is not conditioned
by the transitive valence, but by individual, relevant features of transitivity. Hence,
these alternations differ from each other in semantic transitivity. The basic function
is, however, the same in all cases. Illustrative examples of this include the
following, cf.

(561) tuntuva-k nere-sciu-llru-u-q (carayag-mun)

moose-ABS eat-PASS-PAST-INTR-3SG (bear-OBL)
‘The moose was eaten (by a bear)’
(562) keme-k ner-narq-u-q (yug-nun)
meat-ABS eat-PASS-INTR-3SG (person-OBL)

68
M ore typical in this case are generic man-constructions, e.g. man dansar där, I thank Michaela
Pörn for this remark.

69
The number of participants is naturally the most important feature of transitivity, but the function
expressed is the same in both cases and is not sensitive to any significant feature of transitivity.
 243

‘Meat can be eaten by people’

(563) tauna ner-nait-u-q (yug-nun)
this.ABS eat-PASS-INTR-3SG (person-OBL)
‘This one cannot be eaten (by people)’ (Yup’ik, Payne 1997: 207f, cited from Reed et al
1977)

(564) ma Ø-tzeeq’a-at cheep (t-u’n- kyel)

PAST 3SG.S-hit-PASS 1 José (3SG-REL/AGENT Miguel)
‘José was hit (by Miguel) (on purpose)’
(565) ma Ø-tzeeq’a-njtz cheep (t-u’n Kyel)
PAST 3SG.S-hit-PASS 2 José (3SG-REL/AGENT Miguel
‘José was hit accidentally (by Miguel)’ (Mam, Dixon & Aikhenvald 1997: 75)

(566) bayi bargan ba õgul ya a-õgu `urga u

ABS wallaby ERG man-ERG spear.PRES
‘Man is spearing wallaby’
(567) bayi ya a `aban-du waga õanu
ABS man.ABS eels-ERG(INSTR) spear.õay
‘Man is (actually) spearing eels’
(568) bayi ya a `aban-du wagaymari u
ABS man eels-ERG(INSTR) spear.riy
‘Man is about to spear eels/has speared eels’ (Dyirbal, Dixon 1972: 65, 91)

(569) no-pa-mate-‘e na kene-su mohoo iso

3R-CAUS-dead-3OBJ NOM friend-1SGPOSS sick yon
‘He killed my sick friend’
(570) no-hoko-mate-‘e na kene-su mohoo iso
3R-FACT-dead-3OBJ NOM friend-1SG.POSS sick yon
‘He killed my sick friend (permanently)’ (Tukang Besi, Donohue 1999: 206)

(571) k ‚-ka k‚ namcha-rµl talli-ke ha-ycss-ta

3SG-NOM DET man-ACC run-CAUS do-PAST-DECL
‘He made the man run’
(572) k ‚-ka k‚ namcha-eke talli-ke ha-ycss-ta
3SG-NOM DET man-DAT run-CAUS do-PAST-DECL
‘He let the man run’ (Korean, examples courtesy of Hak-Soo Kim)

(573) a ÷e boe e ÷vd vwa-d f

you NEUT people they know-CAUS
244

‘You caused the people to know’

(574) a-÷e boe e ÷vd vwa-g f
you-NEUT people they know-INCEPTIVE.CAUS
‘You caused people to begin to know’ (Boróro, Rodrigues 1999b: 196)

(575) yara ya-ka-ta õ-kra-t-akn

tree V PL V PL T-1SG A-COM(APPL)-cut-PERF-3SG D
‘I cut his trees for him’ OR ‘I cut his trees’ (negative effect on him)
(576) yara ya-ka-kra-õa-t-akn
tree V PL V PL T-1SG A-cut-BEN(APPL)-PERF-3SG D
‘I cut the trees for him’ (Yimas, Foley 1991: 310)

In the examples above, the basic function, be it passive, antipassive, causative or

applicative is the same, but there are clear transitivity-related differences between
the given alternations.70 In Yup’ik, there are three different passives. Affix -sciu- is
related to actual (affirmative?) events, whereas -narq- and -nait- express affirmative
and negative potential events, respectively. Hence, the events profiled by these
different passives differ in their semantic transitivity. Similarly to Yup’ik, Mam has
two morphologically distinct passive affixes used depending on the degree of
agency involved: (564) denotes a volitional event and (565) an accidental one. In the
latter case, the degree of agency is clearly lower, which also affects the semantic
transitivity of the clause. In languages like English, this difference is expressed by
adverbials. Examples from Dyirbal illustrate two different antipassives (õay and riy-
forms). These affixes differ from each other in the concreteness of the events
denoted. (566) and also (567) describe actual, concrete actions, whereas (568)
describes rather a potentiality. The latter is more appropriate, if a man has gone out
to spear some eels, but we are unaware of whether he has speared any or whether
he is spearing some at the moment. Of the two antipassives, (566) can be considered
more transitive. The Tukang Besi and Korean examples illustrate different kinds of
causative that can be distinguished on the basis of the degree of transitivity. In
Tukang Besi, the decisive factor is the permanence of the effect on the patient. (569)
is appropriate, if someone is dead for a while, but comes back to life afterwards.

70
Unfortunately, in some cases I can show only the resulting structures, but the glosses should suffice
to show that we are dealing with transitivity alternations.
 245

According to Donohue (ibidem), this seemingly bizarre clause can be used, if

someone has been made dead for a while by a shaman, but it is very likely that the
person will come back to life (and perhaps is already alive when the clause is
uttered). Example (570) refers to an event resulting in a ‘permanent’ death. The
same causatives can also be used to signal differences in the degree of affectedness
of the patient (see Donohue 1999:207). The Korean examples are very typical
examples of differences between direct and less direct causation. In (571), the
causation is direct, whereas in (572) we are dealing with a less direct (e.g. purely
verbal) causation. This differences is reflected in the marking of the causee: in direct
cases the causee is the accusative, while in less direct cases it appears in the dative.
A similar variation is attested in numerous other languages as well, including
Finnish, Japanese and Hungarian. The difference between (573) and (574) lies in
whether the causation results (or has resulted) in a change-of-state in the patient or
whether the process at issue has only been initiated (as in (574)). The event denoted
in (573) is more transitive, since it has a definite and salient result. The last two
examples illustrate different applicatives. The main difference here is the nature of
the effect on the indirectly affected participant. In (575), the event has a negative
effect on the indirectly involved participant, while in (576) this is not implied and
the given participant can be regarded as a beneficiary. Furthermore, in (575), the
indirectly involved participant can be viewed as more affected than in (576) The
basic event remains the same. These differences are less directly related to
transitivity than, for example, those in (569) and (570), since the different degree of
affectedness is not a feature of the primary target of the action (i.e. ‘trees’).
Examples (561)-(576) illustrate cases in which the specificity is due to some
relevant transitivity feature. Languages that do not have different alternations
markers for these kinds of alternation usually express these differences somehow
(e.g. by using adverbials).
The last subtype of type 3 is illustrated by cases in which the semantic features
inhering in the markers are not related to any feature of transitivity. A couple of
somewhat different examples of this are illustrated below, cf.

(577) tepau-a-§a
break-3SG-IND
246

‘He broke it/it broke’

(578) tepau-fe-a-§a
break-TR-3SG-IND
‘He broke it/them’
(579) tepau-fu-a-§a
break-TR(PL)-3SG-IND
‘He broke them’ (Tauya, MacDonald 1990: 179, 197)

(580) tsewma õ=ni§ door= §a § §a-ka-kal-piak

Tsewmang=ERG market=ALL/LOC 3SG.S-1SG.O-go-BEN
‘Tsewmang went to the market for me’
(581) thiõ §a-ka-laak-tse §m
wood 3SG.S-1SG.O-carry-ADD BEN
‘He carried wood for me (in addition to carrying wood for himself)’
(582) ka-law §an-ka-thlo §-pii
1SG.POSS-field 3PL.S-1SG.O-hoe-COM
‘They hoed my field (together) with me’
(583) rul=ni§ ka-§in= §a § §a-ka-lu §-hno §
snake=ERG 1SG.POSS-house=ALL/LOC 3SG.S-1SG.O-enter-MAL
‘A snake came into my house on me’
(584) booi §a-ka-toon-ka §n
chief 3SG.S-1SG.O-meet-PRIOR
‘He met the chief ahead of/before me’
(585) §a-law §a-ka-thlo §-taak
3SG.POSS-field 3SG.S-1SG.O-work-RELINQ
‘He left me and hoed his field’
(586) tiiloo õ khaa tivaa kan-Ø-tan-naak
boat TOP river 1SG.S-3SG.O-cross-INSTR
‘W e used the boat to cross the river’ (Haka Lai, Peterson 1998: 36f)

In Tauya, the form of the causative affix is sensitive (among certain other things) to
the number of the Patient (an identical case is found in Embera languages, see
Mortensen 1995:63). Example (578) is ambiguous as regards the number of the
Patient. This ambiguity can be resolved by using the marker -fu- which is restricted
to plural Patients. The feature ‘plural’ does not correlate in any direct way with
semantic transitivity, but the events described by (578) and (579) are ontologically
 247

equally transitive.71 In both cases, the agent is acting volitionally and the patient is
directly affected. Haka Lai is a language with seven morphologically distinct
applicatives. The basic function is shared by all of these: a new argument (referring
to a peripheral participant) is added to the clause core. In a broad sense, (580)-(583)
can be considered slightly more transitive than the rest of the illustrated cases, since
in these cases the applicativized argument refers to an (indirectly) affected
participant (cf. 5.3.1.2.1. below), whereas in (584)-(586), the additional feature is
not related to transitivity in any way. In general, however, there are no significant
differences in transitivity in these cases.
Specific alternations belonging to type 4 are the most obvious examples of
specific transitivity alternations. They are frequently attested in the languages of the
world, very likely in some form in all of them. The difference between changes in
valency and some minor transitivity features are very obvious and it is very natural
that these kinds of alternation are distinguished also structurally. There are, for
example, numerous languages that have morphologically distinct passive and
Involuntary agent constructions (IAC) or distinct experience constructions and
anticausatives. The distinction between these kinds of event is almost as
fundamental as that between valency increasing and valency decreasing alternations.
The far more interesting examples of this type were illustrated above, since the
general markers illustrate the ‘marked case’. One example suffices here, cf.

(587) mies tul-i tappa-nee-ksi karhu--n

man.NOM come-3SG.PAST kill-PAST-TRANSL bear-ACC
‘The man happened to kill the beat’
(588) karhu tape-tt-iin
bear.NOM kill-PASS-PAST
‘The bear was killed’ (Finnish)

In Finnish, IAC’s and passives are distinct constructions. In the former the number
of participants is retained, while the latter results in a complete de-transitivization.
The last type (type 5) of specific alternation is illustrated by cases in which

71
Plural patients can be regarded as less definite, but this is ignored here.
248

minor changes in transitivity are marked linguistically. These kinds of alternation

are also very easy to find, cf.

(589) zamara.di get’e xa-na

Zamira.ERG pot break-AOR
‘Zamira broke the pot’
(590) zamara.di-waj get’e xa-na
Zamira-ADEL pot break-AOR
‘Zamira broke the pot accidentally’ (Lezgian, Haspelmath 1993: 292)

In Lezgian, the ergative marking of Agent is related primarily to high agency,

whereas the adelative signals decreased agency (and transitivity). The shift from
ergative to adelative does not seem to be related to any other transitivity alternations
in Lezgian, because of which we can justly claim that we are dealing with a specific
agency-conditioned alternation in (590).

5.2.3.3. Concluding remarks

In the two previous sections, we have illustrated two different alternation types
labelled as general and specific alternations. These are distinguished by the number
of functionally different alternations expressed, which is closely associated with the
semantic specificity. In the case of general alternations, a single alternation marker
may express a variety of functions the exact nature of which is inferred from the
context. As opposed to this, the nature of an alternation inheres in the semantics of
specific alternations. This makes it impossible for specific alternations to occur in
environments that contradict their semantics. The question, an alert reader might be
asking, is what the typology illustrated above can possibly tell us about the nature
of transitivity? On the next couple pages we try to answer this question.
To begin with, we may say that the illustrated typology enables us to
understand what is a possible alternation. The specific alternations are more
important in this respect, since they are directly related to certain functions. They are
determined by non-linguistic differences that have to (or can) be distinguished. If
two events are sufficiently distinct, it is very natural that the language has the means
 249

to distinguish between them explicitly. In this case, structural differences simply

reflect non-linguistic differences between events, which reveals important facts
about transitivity. We may say that the number of specific alternations conditioned
by specific transitivity parameters coincides with the number of possible transitivity
alternations. One problem we might have to face here is that also many of the
alternations that have here been labelled as specific may have other functions as
well. On the other hand, general alternations aid us in seeing how semantics and
structure co-operate. Many different alternations can be expressed using a single
marker, which implies that linguistic distinction is not necessary. The illustration
above shows that many different alternations can be expressed by a single marker,
which does not result in true ambiguity. The existence of this kind of alternation is
direct evidence for the combinatory nature of transitivity. Linguistic transitivity is
not merely an iconic reflection of non-linguistic transitivity, but many differences
can be neutralized.
The nature of general alternations attested typologically speaks for the last
point made above. The most typical general alternation types are illustrated by types
two and three. In these cases, the functions are very similar to each other and certain
facets of events (or the context) aids us solving the possible ambiguity. For example,
in (591)-(594) the semantics of events and the nature of the sole argument tell us
that only (591) and (593) illustrate reflexives, while (592) and (594) exemplify a
passive or an anticausative, cf.

(591) kaya-Ø kipa-kali-Ø

father-ABS shave-REFL-P/P
‘Father is shaving himself (or someone)’
(592) yuri-Ø watyu-kali-n
kangaroo-ABS cook-REFL-P/P
‘Kangaroo is cooked’ (W arrungu, Tsunoda 1988: 638)

(593) mies pese-yty-i

man.NOM wash-MIDDLE VOICE/REFL-3SG.PAST
‘The man washed himself’ (*man was washed)
(594) kaupunki tuho-utu-i
city.NOM destroy-ANTIC/MIDDLE VOICE-3SG.PAST
250

‘The city was destroyed’ (Finnish)

As illustrated, in Warrungu the reflexive marker can also mark passive. In (592), a
genuine reflexive interpretation is not possible in light of the semantics of the event
and its sole participant. In Finnish, anticausatives and reflexives can be expressed
by the same affix -UtU-. It is unnecessary to have specific markers for these, since
the inanimacy of the referent of the sole argument excludes reflexivity. We know
that only animate entities are capable of volitional actions targeted at themselves.
Hence, a reflexive interpretation is excluded in (592) and (594), while it is the only
plausible one in (591) and (593). Also the inherent semantics of events is relevant
here. Cooking is an event that is instigated exclusively by humans and it is rarely (if
ever) targeted at the instigator itself. Washing, on the other hand, is an inherently
transitive event related to a high degree of agency. Furthermore, the reflexive is the
only possible interpretation, since the event involves obligatorily two semantic roles
which happen to be properties of a single participant in (594).
Specific alternations can aid us in understanding the rationale behind
individual transitivity alternations. Most typically, specific alternations occur, if the
semantic differences between two events are so significant that we have to resort to
linguistic cues in interpreting clauses. For example, every language has some means
to express causativization and reflexivity. The differences between these and basic
transitive events are very obvious, in addition to which the alternations differ from
each other in so many relevant aspects that it is not possible to express them clearly
by a single polysemous marker. Hence, languages employ different mechanisms to
express them. Depending on the language, closely related functions may also be
expressed by specific markers or the language may employ a single marker with
multiple functions. Conversely, if two alternations are semantically very similar, it
is most uneconomical to employ specific markers to express these alternations. For
example, we do not expect to find a language that has two different causatives that
are used according to whether the agent is looking at the patient when performing
an action or not.72 This feature is totally irrelevant for our interpretation of

72
W hat makes this example justified is that this is one of the relevant parameters of a ‘prototypical
agent-patient clause’ as defined by Lakoff (1977:244, cf, above ch. 2.4.).
 251

transitivity and there is no need to subclassify, for example, causatives accordingly.

There are, though, as illustrated above, specific alternations that cannot be separated
from each other by referring to any relevant facet of transitivity. In these cases the
existence of the specific alternations can, however, be explained by referring to
other semantically relevant features. For example, the nature of the introduced
argument is an integral part of applicatives, which explains the existence of
constructions like (580)-(586). This is not to say that transitivity is always irrelevant
for applicatives, since there are applicatives that are used to emphasize a higher
degree of patient affectedness. For example, in Bantu applicatives like ‘sit in a
chair’ are grammatical, whereas the likes of ‘sit in a room’ are not. This variation
follows from the implausible affectedness reading in the latter case.73 On the other
hand, whether the agent is looking at the patient is not a property that would be
lexicalized as a part of the meaning of certain alternations. In this respect, specific
alternations also reflect the economy principle of language, since only alternations
relevant for our conceptualization of events are explicitly distinguished.
Furthermore, since specific alternations are directly related to individual
alternations they enable us to see what kinds of alternation are (possibly)
conceptually different. Since there are many languages that distinguish, for example,
between closely related reflexive and anticausative functions, it is justified to label
these as distinct transitivity alternations. This distinction is naturally semantically
determined. In similar vein, we cannot label the two kinds of causative discussed
above as distinct transitivity alternations, since no language (to our knowledge)
distinguishes between them explicitly. In consequence, specific alternations aid us
in distinguishing different transitivity alternations from each other. The contribution
of general alternations is very different in this respect. Since certain alternations can
be expressed by a single marker, general alternations do not reveal much about
differences between alternations. At best, they aid us in seeing what kinds of change
result in a transitivity alternation in general. As a result, general alternations reveal
what is common for certain alternations. The rareness of type 1 is ‘hard evidence’
for the fact that not a mere change in transitivity suffices to mark transitivity

73
I thank Peter Austin for this remark.
252

alternations by a common marker, but they usually have to share certain features in
order that a general alternation marker can be used. The types of general alternation
illustrated above differ from each other in how the generality is motivated. In the
three first types the different alternations are ‘equal’ in the sense that the use of an
alternation marker is enabled by a function common for all the alternations marked
identically. For example, in the case of the commonly attested structural polysemy
of reflexive (and in many cases also middle voice), anticausative, passive and
reciprocal this common feature is the omission of an argument. Also the semantic
role of arguments is common for these constructions. There are clear semantic
differences as well. For example, events profiled by passives and reciprocals involve
two participants, whereas in anticausative and reflexives the syntactic omission of
arguments coincides with a semantic omission of participants. The basic
de/intransitivizing function is, however, the same. The other possibility is that the
generality is motivated hierarchically. In this case, one of the alternations is more
important than others, since it expresses all (or at least most of) the possible features
associated with an alternation type. This is most obvious in type 5. For example, in
Finnish the nominative-partitive frame is associated generally with a lower inherent
transitivity of events. The shift from an accusatively to a partitively marked Patient
produces a semantic change in the roles of both participants. The change in (530)
is not related to both participants, but only the affectedness parameter is influenced.
The changes in the inherent semantics are more holistic and other alternations only
share some features with it.
The five types illustrated above differ from each other in whether it is more
natural to express the alternations in question by general or specific alternations. As
implied above, we expect the specific alternations to outnumber the general ones in
type 1, since the differences between the alternations expressed are very obvious.
The alternations do not have common functions nor can we say that the alternations
at issue stand in a hierarchical relation to each other. Hence, our findings are very
natural. Type 2 is clearly the most heterogeneous of the illustrated types. The basic
function and the nature of the alternation is the same in all cases, but there are also
obvious differences. The first of these properties enables the use of general
alternations, since, as illustrated above, the exact nature of the alternations is
 253

inferrable from the semantics of events and participants involved. On the other
hand, the subtle differences make it also possible to distinguish the alternations from
each other explicitly. This heterogeneity is also reflected in actual linguistic data.
General and specific alternations are more or less equally distributed. We are not
surprised to find either types as regards, for example, the expression of
anticausative, reflexive and passive, since the existence of them both is easily
accounted for. The third type is exceptional. It is justified to claim that it is more
natural to use general markers in this case. The semantic differences between
individual alternations are rather insignificant and employing distinct alternations
obviously violates the economy principle of language. It is, for example, not
necessary to have distinct causative affixes for intransitive, transitive and
ditransitive clauses, since the function expressed is the same. Furthermore, these
different instances of causativization are semantically very close to each other,
which makes distinct markers seem redundant. The use of different alternation
markers is more readily understandable, if the semantic feature at issue is somehow
related to transitivity, as is, for example, in (569)-(574). Type 4 emphasizes the
more general nature of transitivity and the correlations of different ‘levels’ with each
other. In this case, the existence of specific alternations does not have to be justified
in any way. Changes in transitive valence and in individual transitivity features
illustrate very different alternations and it should not come as a surprise that the
majority of languages distinguish these alternations explicitly. These alternations
share, however, some common features and there are languages that employ general
markers in these cases. For example, the causative affix is also used to intensify or
agentivize transitive events in genetically unrelated languages, such as Navajo,
Chichewa, Godoberi and Kambera. This kind of polysemy becomes obvious, if we
take account of the semantics of causatives. Causativization introduces an agent to
an event. Hence, it is natural that the same alternation marker can also be used to
agentivize transitive events that lack a genuine agent. It seems, however, that
valency affecting and ‘other’ alternations are usually kept apart, since general
markers in cases like (503)-(522) seem to be less frequent than specific markers.
Specific markers are the default choice here. The fifth type is also very
heterogeneous, and the alternations in question can be marked with both general and
254

specific alternations.

5.2.4. Transitivity increasing, transitivity decreasing and transitivity rearranging

alternations

Transitivity alternations can be divided roughly into three on the basis of their
structural consequences. These are here referred to as transitivity increasing,
transitivity decreasing and transitivity rearranging alternations. The distinction is
here based on whether the alternations affect the number of arguments or whether
they only impinge upon the marking of arguments. The two first alternations
exemplify the former type, whereas the last type illustrates the latter kind of
alternation. As the labels imply, two first types differ from each other in whether
they increase or decrease the number of arguments expressed. In the following, we
will illustrate and discuss the types primarily from a structural perspective without
ignoring semantics. We divide the alternations into different types according to the
underlying semantic motivation and on the basis of their consequences for the
structure of events. The alternations are illustrated in the order they appear in the
title.
What is important for transitivity (or valency) increasing alternations is
naturally that the number of arguments in clauses increases. The transitivized
clauses involve an argument that is lacking in the less transitive clauses. It is
irrelevant whether the increase in the number of arguments also increases the
semantic transitivity of clauses or not. Hence, a more plausible term could be
‘valency increasing alternations’. The increase may or may not be accompanied by
morphological changes in the verb. Typical examples of transitivity increasing
alternations are the following, cf.

(595) k hic’ q’ena

dog.NOM died
‘The dog died’
(596) gada-di k hic’ q’ena
boy-ERG dog.NOM died
‘The boy killed the dog’ (Lezgian, Job 1985: 161)
 255

(597) o-ne k’aNa k’ad-a

3SG-ERG food eat-PAST.MSG
‘He ate food’
(598) timi-ne o-nu k’aNa k’lay-a
woman-ERG 3SG-DAT food eat.CAUS-PAST.MSG
‘The woman made him eat food’ (Panjabi, Payne 1997: 180, examples courtesy of Lynn
Conver)

(599) taveti-a mosi sa lose

make.TR-3SG Mosi ART.SG room
‘Mosi built the room’
(600) tavete-ni-a kinahe mosi sa lose
make-APPL-3SG nipa.palm Mosi ART.SG room
‘Mosi built the room with nipa palm’ (Hoava, Davis 1997: 239)

(601) nganaju yaan (pawulu-ngara-a) kampa-lalha murla-a

1SG.GEN wife (child-PL-ACC) cook-PAST meat-ACC
‘My wife cooked meat for the kids’ (Martuthunira, Dench 1995: 68)

(602) isä luk-i kaappokirja-n jaako-lle

father.NOM read-3SG.PAST Kaappo.book-ACC Jacob-ALL
‘Father read the book about Kaappo to Jacob’ (Finnish)

Examples from Lezgian and Panjabi illustrate causatives. In Lezgian (in this
particular case), the verb is not marked, but the primary difference between (595)
and (596) lies the number of arguments. Examples from Panjabi illustrate a
transitive causative derivation from signalled also on the verb. These are genuine
examples of transitivity (or valency) increasing alternations, since the added
argument refers to an additional participant that contributes to the nature of the
event denoted as well. Examples (599)-(602) illustrate different applicative
constructions. The label is used here very loosely and any construction with an
added argument that does not refer to an agent or a causer is regarded as an
applicative. The Hoava examples illustrate a typical applicative in which the adding
of the argument is accompanied by morphological changes in the verb, which
promotes the added argument to the core. Examples from Martuthunira and Finnish
exemplify cases in which an argument is introduced to a transitive clause without
256

this being marked on the predicate. The two languages differ from each other in the
marking of the added argument. In Martuthunira, the beneficiary appears in the case
of typical Patients, while in Finnish it appears in the allative case. The added
argument is completely optional (and can hence also be labelled as an optional
adjunct). Also in Hoava, the added argument refers to a peripheral participant
unrelated to semantic transitivity. However, the status of the added argument
distinguishes between these two alternations. Only in Hoava are we dealing with a
core argument and only these examples qualify as genuine valency increasing
alternations. The semantic consequences for the transitivity of events are irrelevant,
since syntax overrides semantics. The likes of (601) and (602) are not ignored either,
but their status is considered somewhat lower.
As is evident, transitivity decreasing alternations illustrate the exact opposite
of transitivity increasing alternations. This means that transitivity decreasing
alternations reduce the number of arguments in clauses. Similarly to transitivity
increasing alternations, this may, but need not, be accompanied by morphological
changes in the verb. Typical examples of functionally different transitivity
decreasing alternations include the following, cf.

(603) mawun-du ga `i `ilmu man-i

man-ERG tree break-PAST
‘A man broke a tree’
(604) ga `i `ilmu ira
tree break-PAST
‘A tree broke’ (Djaru, Tsunoda 1987: 192)

(605) azniv-c tur-c kots-ets (*kotsvetsav)

Aznive-the door-the close-AOR/3DG
‘Aznive closed the door’
(606) tur-c kots-v-ets-av
door-the close-INTR(v)-AOR-3SG
‘The door closed’ (Modern W estern Armenian, Haig 1982: 164f)

(607) nídbá nyìrá dáám wèém

people.PL drink.PFV corn.beer quickly
‘The people drank the beer very quickly’
 257

(608) dáám nyìrá wèém

corn.beer drink.PFV quickly
‘The beer was drunk quickly’ (Nawdm, Watters 2000: 211, examples courtesy of Jacques
Nicole)

(609) er schlug den mann

he.NOM hit.PAST.3SG ART.ACC man
‘He hit the man’
(610) der mann wurde (von ihm) geschlagen
ART.NOM man become.PAST.3SG (‘by’ he.DAT) hit.PARTIC
‘The man was hit (by him)’ (German)

(611) ngamwa n-aa-shereka ekyanzi

PN PC-he-hide milkpot
‘Ngamwa is hiding the milkpot’
(612) ngamwa n-aa-ye-shereka
PN PC-he-self-hide
‘Ngamwa is hiding (himself)’ (Nkore-Kiga, Taylor 1985: 65)

(613) mies sö-i (leivä-n)

man.NOM eat.PAST-3SG (bread-ACC)
‘The man ate (the bread)’ (Finnish)

(614) na’e inu ‘a e kavá ‘e Sione

PRET drink ABS the kava ERG John
‘John drank the kava’
(615) na’e inu (kavá) ‘a Sione
PRET drink (kava) ABS John
‘John drank (kava)’ (Tongan, Churchward 1953: 76)

(616) ebel-a 8 ret-el b-úq §-ula

mother-ERG clothes.ABS-PL AGR.ABS-sew-PRES
‘Mother sews clothes (in general)’
(617) ebel y-uq §-ár-ula
mother.ABS AGR.ABS-sew-ANTIP-PRES
‘Mother does (some) sewing’ (Avar, Tchekhoff 1985: 375)

What is important for our purposes is that the number of explicitly expressed
258

arguments has been reduced by one and the result is always a syntactically
intransitive clause.74 Whether this is indicated in the verb is irrelevant. The nature
of omitted arguments is different in different cases. In (603)-(610) the eliminated
argument refers to an agent, whereas in the rest of the examples the Patient is
omitted. The examples differ from each other also in how the omission is motivated.
One important facet of this is whether the omission is semantically motivated or not.
If a participant is eliminated altogether from an event, it is naturally impossible to
refer to it. Examples (603)-(606) illustrate this. However, there are clear differences
between the examples in how the semantic omission is reflected on the level of
morphosyntax. In (604), the Agent omission is not signalled on the verb.
Consequently, the expression of agent is structurally possible, even if the semantics
of the event excludes the agent. Examples like these are rather problematic as
regards the exact nature of the alternation in question. They are structurally identical
to (595) and (596) from Lezgian and it is very difficult (if not impossible) to say
whether we are dealing with a transitivity increasing or with a transitivity decreasing
alternation. In (605) and (606), the omission of the agent is accompanied by an
intransitivization of the verb, which makes the expression of Agent also structurally
impossible. Examples (607)-(610) illustrate passives that are similar to (603)-(606),
since the Agent is omitted. In (608), the passive is derived simply through an Agent
omission. It is possible to re-introduce the Agent to (608), if so desired. In (610)
from German, the passive is derived by manipulating the verb morphology, which
results in changes in the argument marking as well. The passive morphology of the
verb does not exclude the expression of Agent, but it has to be added in an oblique
form. Examples (611) and (612) illustrate a reflexive. On the basis of the number
of participants, reflexives are also intransitive constructions. Consequently, it is not
a surprise that structurally intransitive reflexives are quite typical in the languages
of the world. Nkore-Kiga exemplifies this kind of language. In Nkore-Kiga, the verb
is not intransitivized, since the reflexivity is expressed by attaching a reflexive affix
to the verb. This intransitivizes the clause, and it is no longer possible to express the
Patient as an independent argument in (612). Examples (613)-(617) illustrate

74
Ditransitives are excluded in the present context.
 259

different antipassives (defined in a rather loose sense). In (613) from Finnish, the
Patient is simply omitted without any further structural changes. It is, therefore,
possible to ‘re-introduce’ the Patient to the clause. Structurally (and only
structurally), the Patient can be said to be an optional argument (cf. e.g. Lehmann
1991: 192). It is very obvious that we are dealing with a transitivity decrease in
(613), since the argument omitted refers to a core participant. In Tongan, the Patient
is omitted without this resulting in any morphological changes in the verb.
Differently from Finnish, however, there is a change in the marking of the argument
referring to the agent. In (614) (i.e. in a transitive clause), the Agent appears in the
ergative, whereas in (615) the Agent is in the absolutive. This change is due to the
basic ergative structure of Tongan. In Avar, the Patient omission is also signalled
on the verb, i.e. (617) illustrates typical antipassive construction.
The two alternation types illustrated above differ crucially from each other as
regards the kinds of omission that qualify as transitivity alternations. As illustrated
in (580)-(586), for example, the introduction of semantically various arguments can
result in a transitivity increase. On the other hand, typically only cases in which an
argument referring to a concrete participant in an event result in a genuine
transitivity decreasing alternation. Put concretely, this means that semantically only
cases like ‘he killed him’ vs. ‘he died’ are instances of transitivity decrease, whereas
the likes of ‘he sang in the yard’ vs. ‘he sang’ are not. The location is not an integral
part of ontological transitivity. The rationale behind this difference is that it is
impossible to omit a participant (or rather a circumstant) that is not an integral part
of an event. In order that the omission is genuinely possible, we have to add the
argument at issue first. Furthermore, this kind of omission is not motivated by any
aspect of transitivity. At the level of morphosyntax, these two features are reflected
in the fact that there are (very likely) no languages in which this kind of omission
would be accompanied by morphological changes in the verb. Seeming examples
of this are, however, illustrated below, cf.

(618) kom pe ina-in nahwarak

sing be.at.SG 1SG.AG.REAL-3SG.N.PAT jungle
‘I sang in the jungle’
260

(619) kom ina-in chek ka § ne

sing 1SG.AG.REAL-3SG.N.PAT day this.N REC.PAST
‘I sang this morning’
(620) kom ina
sing 1SG.AG.REAL
‘I sang’
(621) to § na-on (>non) narima § tarama §
hit 3SG.AG.REAL-3SG.M.PAT man woman
‘The woman hit the man’
(622) to § na
hit 3SG.AG.REAL
‘S/he is/was hitting’ (W ari)

Examples (618), (619) and (621) are transitive, whereas (620) and (622) exemplify
intransitive constructions. The main mechanism of ‘intransitivization’ is the
omission of arguments which also omits the reference to the given argument in the
person clitic. Examples (620) and (622) are structurally similar to each other in this
respect, but the semantic differences are rather obvious. In (620), the omitted
argument is a circumstant, while in (622) we are dealing with the omission of a core
argument. Hence, only in (622), an integral part of the event is omitted. Since the
result is the same, someone may be led to think that (620) illustrates a case in which
the omission of a circumstant results in a genuine transitivity decreasing alternation.
However, (618) and (619) illustrate cases in which a circumstant is referred to in the
person clitic. Typically, the Patient slot is occupied by a true Patient, as in the
following, cf.

(623) kerek pe §ina-on orowao pain nahwarak

see be.at.SG 1SG.AG.REAL-3SG.M.PAT Orowao PREP jungle
‘I saw Orowao in the jungle’

If there is a genuine Patient present, it is not possible for a circumstant to occupy the
Patient slot in the clitic. Consequently, (618) and (619) rather illustrate a kind of
applicative and they are best regarded as transitivity increasing alternations.
The examples illustrated so far have illustrated cases in which either an
‘external’ argument or participant is introduced or in which arguments are
 261

completely eliminated. These are perhaps the most typical examples of transitivity
increasing and decreasing alternations. Before proceeding to transitivity rearranging
alternations, we will illustrate two transitivity alternations, the former of which is
a kind of subtype of the increasing type, while the latter is related to the decreasing
type. What entitles us to view these alternations as distinct types is that nothing
(semantically) external is added nor is anything completely eliminated. These cases
rather illustrate alternations that have very radical consequences for the status of
arguments (or participants). The constructions under study are external possession
(here only the ‘European type’, cf. Haspelmath 1999), body part ascension
alternations (cf. Levin 1993: 149), cognate object constructions, and noun
incorporation. The two first are here understood as subtypes of the same alternation
type, since their effect on the structure of clauses is more or less the same. All of
these are illustrated below, cf.

(624) ich wasch-e mein-e hände

I.NOM wash.PRES-1SG 1SG.POSS-PL hands
‘I wash my hands’
(625) ich wasch-e mir die hände
I.NOM wash.PRES-1SG I.DAT ART.PL hands
‘I wash my hands’ (German)

(626) sus ojos se llenaron de lágrimas

his eyes REFL filled of tears
‘His eyes filled with tears’
(627) los ojos se le llenaron de lágrimas
ART eyes REFL he.DAT filled of tears
‘His eyes filled with tears’ (Spanish, Haspelmath 1999: 112, cited from Roldán 1972: 27)

(628) hän läimäytt-i jaakko-a

s/he slap-3SG.PAST Jacob-PART
‘S/he slapped Jacob’
(629) hän läimäytt-i jaakko-a selkä-än
s/he slap-3SG.PAST Jacob-PART back-ILL
‘S/he slapped Jacob on the back’ (Finnish)
262

(630) nganhi (diyari yawada) yatha-yi

I.NOM (Diyari language.ABS) speak-PRES
‘I speak Diyari’ (Diyari, Austin 1982: 39)

(631) subih t ]b vvh ta-ma

Subih house 1SG THEME-make
‘I am making Subih 's house’
(632) subih vvh t]b-ta-ma
Subih 1SG house-THEME-make
‘I am making a house for Subih’ (Nadëb, Martins & Martins 1999: 261)

(633) ø-qui-cua in pilli in nacatl

3SGI-3SGII-eat ART child ART meat
‘The child eats the meat’
(634) ø-qui-cua nacatl in pilli
3SGI-3SGII-eat meat ART child
‘The child eats meat’
(635) ni-naca-cua
1SGI-meat-eat
‘I eat meat’ (Classical Nahuatl, Lazard 1998: 187, cited from Launey 1979: 40, 166)

In (624)-(630), the number of independent arguments increases. In the three first

cases the number of arguments increases from two to three, while in (630) the
increase is from one to two. Examples (624)-(627) illustrate very typical external
possession constructions in which the possessor has been detached from the subject
argument and it appears as an independent argument in the dative. In (629) from
Finnish, a body part of the targeted participant is focussed on, which is reflected in
the grammatical status of the argument as well. As opposed to cases illustrated in
(603)-(610), nothing new as such is introduced to the events profiled. There are no
hands or eyes without a person they belong to. In similar vein, the back is an
intrinsic part of the human body. In (625), (627) and (629), these are explicitly
focussed on. Example (630) is somewhat different from the previous ones. In this
case the ‘product’ of an intransitive action is made explicit. It is common for (624)-
(630) that something implicit is made explicit by referring to it by an independent
argument, which increases the number of arguments. This does not increase the
number of concrete participants, but a subpart of a participant is given more
 263

linguistic prominence. As a result, these kinds of alternation are here distinguished

from alternations that increase the number of participants as well.75 Some of the
alternations that have been labelled as transitivity increasing above share common
features with the alternations illustrated in (624)-(630). The best example of this is
illustrated by applicatives that add an argument that refers to a location. Every event
has to take place somewhere because of which the location is an intrinsic part of
every event. In the case of locative applicatives, the location is focussed on and is
referred to by a core argument. However, applicatives of this kind are excluded here
for two reasons. First, they are treated as a homogeneous transitivity increasing
alternation type in the present context despite their obvious semantic heterogeneity
(cf., however, below). Second, the degree and the motivation of intrinsicness are
different in the two cases. Every event has to take place somewhere, but this is
usually not inferable from any aspect of events. Furthermore, the location is not an
intrinsic part of any of the participants in the event. On the other hand, the fact that
eyes belong to someone and that humans have a back inheres in the nature of the
entities in question. Consequently, in cases like (624)-(630) we can more justly
claim that we are making something truly intrinsic explicit. The two arguments in
cases like (625) or (627) refer to the same participant. Incorporations illustrated
above are in this respect a kind of mirror image of alternations illustrated in (624)-
(630) (cf. also Lehmann 1991: 194). As a result of an incorporation, an argument
loses its status as an independent argument. The derived clause has one independent
argument less than the original one. It would not do justice to the nature of
incorporation to say that we are here dealing with a typical argument omitting
alternation, even if the number of independent arguments decreases. The
incorporated arguments remain parts of clauses. As a result, it is more appropriate
to speak of a drastic decrease in the argument status: incorporated arguments cannot
usually be specified by determiners or adjectives and they are not marked for case
(see, e.g. McGregor 1997: 150). Also semantically, incorporated nouns refer to a
participant. For example, differently from he eats, in which the patient is implicitly
a part of the event denoted, the participant is explicitly referred to in cases like (632)
and (635). The syntactic status of the argument is, however, considerably lower than

75
The nature of participants is irrelevant in this respect.
264

in typical transitive clauses.

The last major type of structural transitivity alternations is exemplified by
alternations that do not affect valency, but only have consequences for the status
and/or the semantic nature of arguments. These are here referred to as transitivity
rearranging alternations. We have adopted the term from Dunn (1999: 215) who
uses the label to refer to alternations like the following, cf.

(636) ctl§a-ta j cme-nenat ewir §-c-t

mother-ERG hang-3SGA.3PLO clothing-E-3PL.ABS
‘Mother hung up the clothes’
(637) ctl§a-ta ena-jme-nen nilp-c-n
mother-ERG APPL-hang-3SGA/3SGO cord-E-3SG.ABS
‘Mother hung (something) on the cord’ (Chukchi, Dunn: ibidem)

(638) arr-iny-mi
1SG.S-2SG.O-get.PP
‘I grabbed you’ (*I got it for you)
(639) arr-iny-nat-mi
1SG.S-2SG.O-BEN-get.PP
‘I got it for you’ (W array, examples taken from a handout given by Adam Saulwick at ALT
IV)

(640) y-ttcu wqqzin aysum

3M S-eat dog.CST meat
‘The dog ate the meat’
(641) y-ss-ttc wryaz aqqzin
3M S-TRANS-eat man.CST dog
‘The man fed the dog’ (Berber, Guerssel 1986: 36)

What is important for the notion of transitivity rearranging in the present context is
that the reference relations of arguments are affected. For example, in Chukchi the
absolutively marked argument refers to a patient in typical transitive clauses, as in
(636). On the other hand, in (637) (that illustrates an applicative construction of
Chukchi) the Patient refers to a location and the patient is not explicitly referred to.76

76
The demoted patient can be referred to explicitly by an oblique argument.
 265

In this case, the location outranks the patient as regards which argument gets to be
explicitly referred to by an independent argument (cf. (623) from Wari). Differently
from typical applicatives, the number of arguments does not increase in the
illustrated case. Examples (638) and (639) from Warray also illustrate an applicative
alternation that results in a change in the semantic role of the Patient. In (638), it
refers to a patient and in (639) to a benefactive. Examples from Berber illustrate a
transitivity rearranging causative alternation. In (640), the nominative (CST)
argument refers to an agent, whereas the zero marked argument is a Patient. In
(641), the distribution of the marking is the same, but the semantic roles referred to
are different. The CST argument refers to the causer of ditransitive causatives, while
the zero marked argument refers to the causee (the patient is not referred to in this
case). There is also a minor change in the semantic role of the CST argument. It
refers in both cases to the participant primarily responsible for the occurrence of
events. However, in (640), it refers to the participant that carries out the given
action, whereas in (641) it refers to an instigator. The consequences of rearranging
are somewhat more radical in this case. Further examples of transitivity/valence
rearranging alternations are provided by dative shifts and ‘supply-verbs’. What is
common for all of these constructions is that the reference relations change.
In the present context, we use the term transitivity rearranging alternation in
a broader sense. This does not mean that we would exclude the likes of (636)-(641),
but our definition comprises more cases. The label refers to every case in which the
transitivity of clauses is affected without any consequences for the number of
arguments. Transitivity in the present context is largely identical to what Tsunoda
(1985: 387f) labels as ‘case frame’. This means that changes in transitivity are
expressed by manipulating the case marking of argument(s). Transitivity rearranging
alternations can be subdivided on the basis of whether the changes are semantically
motivated or not and on whether the marking of one or more arguments is impinged
on. Semantic vs. non-semantic alternations are illustrated in the following, cf.

(642) ngatyu-ngku nga-nra-Ø ngumpirr-Ø pat-man-i

I-ERG AUX-1SGI-3SGII woman.ABS touch-do-PAST
‘I touched the woman’ (action completed)
266

(643) ngatyu-ngku nga-rna-la ngumpirr-la pat-man-i

I-ERG AUX-1SGI-3SGIII woman.DAT touch-do-PAST
‘I tried to touch the woman’ (action attempted) (Djaru, Tsunoda 1994: 4676)

(644) vale-a tadi ria koburu rao sa igana

give-TR-3SG PREP.3PL ART.PL child PRO.1SG ART.SG fish
‘I gave the fish to the children’
(645) vale-ni-a igana rao sa koburu
give-APPL-3SG fish PRO.1SG ART.SG child
‘I gave the child fish’ (Hoava, Davis 1997: 238)

(646) yiõu gur õga õanda ma õga-õ

this.ABS kookaburra I.DAT laugh-PRES
‘This kookaburra is laughing at me’
(647) õa a yi `u:-õ gur õga:-õ ma õga:-õa-l
I.ACC this-ERG kookaburra-ERG laugh-APPL-PRES
‘The kookaburra is laughing at me’ (Yidiñ, Dixon 1977: 307)

The alternation presented in (642) and (643) is motivated by the lower degree of
affectedness and event completion, i.e. the change is semantic in nature. This is
manifested in the morphological marking of the Patient (both on the verb and the
argument itself). In (644) and (645), the alternation is not motivated by any relevant
change in transitivity, because of which the alternation is here labelled as primarily
structural. The effect of the applicativization manifests itself as the promotion of the
prepositionally marked argument to core status. The same alternation type, but
without changes in the verb morphology, is illustrated by the English translations
of the Hoava examples. This alternation represents what is typically labelled as
dative shift. Examples from Yidiñ present a complete transitivization of a
morphologically intransitive clause.77 The case frame shifts from morphologically
intransitive to morphologically transitive.
Djaru and Hoava can also be used to illustrate what can be labelled as
argument demoting and argument promoting alternations. The demotion refers to
the removal of a core argument from the core, whereas the promotion is the opposite

77
Dixon does not discuss the semantic motivation behind this alternation in any way, so we cannot
state whether the alternation is semantically motivated or not.
 267

of this. In (642), the Patient appears in the zero marked absolutive case. As a result
of a change in the semantic role of its referent, the Patient is demoted to the dative.
In (645), an argument in an oblique status (or an adjunct) has been promoted to core.
The former of these two alternation types coincides loosely with transitivity
decreasing type, whereas the latter share common features with transitivity
increasing alternations. The given labels are usually applicable only to cases in
which the marking of one argument only is affected. If the marking of both
arguments is impinged on, one of the arguments is usually promoted and the other
demoted. This is, for example, the case in (647) in which the Patient has been
promoted and the Agent demoted. Consequently, it is not justified to label the
alternation only on the basis of one of the relevant aspects.78 The transitivization as
such is independent of whether a given alternation is a demoting or a promoting one.
For example, in (647) the clause as such is clearly promoted to complete transitivity,
even if it is difficult to label the alternation as either demoting or promoting.
In this section, we have illustrated and discussed three different transitivity
alternations as distinct types. The actual picture is, however, more heterogeneous
than might have been implied. There are numerous clear cases on which our
typology is based. For example, anticausatives are unambiguously examples of
transitivity decreasing alternations. However, many alternation types can be
regarded as instances of more than one type, which naturally poses problems for our
analysis. Obvious examples of these are provided by typical passives and
antipassives that allow the demoted argument to be expressed explicitly. If the
Agent or the Patient is completely omitted, passives and antipassives are best
labelled as transitivity decreasing alternations, whereas they are more similar to
transitivity rearranging alternations, if the given argument is merely demoted.
Consequently, it may feel unjust to label these alternations as belonging exclusively
to either of these types. The following examples from Chukchi and Berber are also
somewhat problematic, cf.

78
Cf. Foley & Van Valin 1985: 306ff who divide passives into foregrounding and backgrounding
ones based solely on whether the Patient is promoted to subject status or not.
268

(648) ctl§a-ta ena-jme-nen t ct cl meni p-e

mother-ERG APPL-hang-3SGA/3SGO door.3SG.ABS cloth-INSTR
‘Mother hung the door with cloth’ (Chukchi, Dunn 1999: 215)

(649) y-ss-ttc wryaz aysum i-wqqzin

3MS-TRANS-eat man.CST meat DAT-dog.CST
‘The man fed meat to the dog’ (Berber, Guerssel 1986: 36)

As illustrated in (648), the applicative construction of Chukchi allows the demoted

original argument to be expressed as well (yet in an oblique form). This poses no
greater problems for our typology, though, since the role of the absolutively marked
argument changes. Examples from Berber are more problematic in this respect.
Example (641) (cf. above) illustrates a transitivity rearranging alternation. The
semantic role referred to by the zero marked Patient is different from that in typical
transitive clauses. However, the nature of the alternation changes radically, if all
three participants are referred to, as in (649). In this case, the Patient refers to the
patient and the Agent to an instigator of the event. As is typical of alternations that
increase the number of arguments, one of the arguments appears in a non-core case
that in (649) is the causee, which is typical of causatives in general. Thus, (649) is
an excellent example of a transitivity increasing alternation. How should we analyze
these cases that can, depending on the context, be analyzed in more than one way?
One possible solution is simply to ignore the possible different realizations of one
and the same alternation type in particular languages and to focus on the types per
se. In this case, it is irrelevant whether the passive of a language can be labelled as
both a transitivity decreasing and transitivity rearranging alternation. Since the
purpose of the present illustration is not to classify languages, but different
alternations, this seems intriguing and also possible. However, this is the easy way
out, since even if we are not proposing a typology of languages, two different
analyses of a single construction still seems unsatisfactory. The best we can do here
is perhaps to admit that there are clear and less clear cases and the latter simply
make the proposed division less satisfactory.
Transitivity decreasing and increasing alternations illustrated above form a
clear continuum based on how obligatory or optional the expression of arguments
 269

is and how this is marked. Transitivity rearranging alternations are excluded from
the discussion here. An important aspect in this respect is whether the omission or
introduction of arguments corresponds to a change in the number of (core)
participants. For example, if a participant is completely eliminated, it cannot be
referred to. However, not alone the semantics, but also the morphology of verbs
makes a contribution to the expression of arguments, as will be shown.
In the first case, the expression of arguments is structurally completely
optional. Two different cases of this are illustrated below, cf.

(650) er iss-t fisch

he.NOM eat.PRES-3SG fish
‘He eats fish’
(651) er iss-t
he.NOM eat.PRES-3SG
‘He eats’ (German)

(652) yo ka bouyi dlo-la

they NONPAST boil water-DEF
‘They are boiling the water’
(653) dlo-la ka bouyi
water-DEF NONPAST boil
‘The water is boiling’ (Lesser French Antillean Creole, Gadelii 1997: 111)

In German, the Patient and in Lesser French Antillean Creole (henceforth LFAC for
short), the Agent can be freely eliminated without any other changes in the structure
of clauses. The only structural differences is that in LFAC the Patient has been
removed from the Patient slot and has been promoted to the subject slot. The verb
morphology and the marking of the remaining argument are not affected. In similar
vein, the omitted arguments can be re-introduced. What distinguishes these two
cases is that in German the argument omission does not have any consequences for
the ontology of the event denoted. The event itself remains transitive irrespective of
changes in the number of arguments. In LFAC, the number of arguments
corresponds in this particular case coincides with number of participants.
Consequently, in LFAC, the optionality is restricted to structure. It is semantically
not possible to refer to the agent, if it has been eliminated.
270

The second type of valency affecting alternation is illustrated by cases in

which the expression of arguments is in principle completely optional, but changes
in the number of arguments result in some minor changes. Examples include the
following, cf.

(654) neš-li gazet-Ø b-u -uli

mother-SG.ERG newspaper-SG.NOM OBJ.SG.NONHUM -read-GER
sa-ri
be-SUBJ.SG.FEM
‘Mother is reading a newspaper’
(655) neš (gazet-li) r-u -uli sa-ri
mother (newspaper-SG.ERG) SUBJ.SG.FEM-read-GER be-SUBJ.SG.FEM
‘Mother is reading ((at) a newspaper)’ (Dargi, Mel’cuk 1993: 34f).

(656) er zerbrach den teller

he.NOM break.PAST.3SG ART.ACC plate
‘He broke the plate’
(657) der teller zerbrach
ART.NOM plate break.PAST.3SG
‘The plate broke’ (German)

The Patient (in Dargi) and the Agent (in German) are in theory optional arguments
in the examples above. In the present context, this means that they can be eliminated
without any other changes in the verb morphology. These are similar to (650)-(653),
since in Dargi the argument omission is purely structural, while in (657) the
optionality is restricted to structure. Differently from (650)-(653), there is also some
transitivity rearranging involved. In Dargi, the zero argument refers to a patient in
typical transitive clauses, whereas in (655) it refers to an agent. The situation is
converse in German. The arguments are more closely integrated into the
constructions, even if the expression is not conditioned by the verb.
In the examples illustrated so far, the verb morphology has been completely
neutral, as regards the argument expression. There are also numerous cases in which
these changes are accompanied by morphological changes in the verb. The
expression of arguments is (to different extents) conditioned by the verb
morphology. Typical cases are illustrated by languages in which (anti)passives are
 271

a morphologically marked voice category. A somewhat ‘less radical’ alternation of

this type is found in Sasak (Austin 2000: 7), cf.

(658) kanak=nó jangke=n pancing *(mpaq/lépang/léndóng)

child=that PRES=3 catch fish/frog/eel
‘The child is catching fish/frog/eels’
(659) kanak=nó jangke=n mancing (mpaq/lépang/léndóng)
child=that PRES=3 N.catch (fish/frog/eel)
‘The child is catching (fish/frog/eels)’

Both (658) and (659) are possibly transitive clauses. These examples can be
distinguished on the basis of whether the expression of the Patient is obligatory or
not (which is due to referentiality; in (659) the Patient must be non-referential, even
if it is explicitly expressed). In (658), the Patient is a structurally obligatory part of
the clause and it cannot be omitted without this resulting in ungrammaticality. For
the purpose of eliminating the Patient, the verb morphology must be manipulated,
as in (659). Instead of a completely optional expression of arguments, the omission
may be considered optional (or controlled) in Sasak. In this respect, Sasak differs
from Finnish, German and English in which Patients are freely omissible.
In Sasak, the optional Patient omission does not result in any changes in the
marking of the only expressed argument, even if the verb morphology is
manipulated. This is different from typical passives and antipassives in which the
omission of arguments usually results in transitivity rearranging, even if this is not
an integral part of (anti)passives. Examples are given below, cf.

(660) yi `u:n bu a:-õ mayi buga-õ

this.ERG woman-ERG vegatables.ABS eat.PRES
‘This woman is eating vegetables’
(661) yiõu bu a buga:-`i-õ
this.ABS woman.ABS eat-ANTIP-PRES
‘The woman is eating’
(662) wagu: `a giba: `i u bu a:nda
man.ABS scratch.ANTIP.PAST woman.DAT
‘The man scratched the woman’ (Yidiñ, Dixon 1977: 279, 274)
272

(663) kamuy umma rayke

bear horse kill
‘A bear killed a horse’
(664) umma (kamuy orowa) a-rayke
horse (bear from) PASS-kill
‘A horse was killed by the bear’ (Ainu, Shibatani 1990: 57)

(665) khay pà § mpò õ

he open door
‘He opened the door’
(666) mpò õ g c-pà § (m c cal)
door PASS-open (by wind)
‘The door was opened (by the wind)/the door opened’ (Sre, Keenan 1985: 252, cited from
Manley 1972)

Examples (661) and (662) illustrate the antipassive of Yidiñ. This alternation results
optionally in a decrease in the number of arguments, as shown above. The (optional)
omission of the Patient is accompanied by a morphological change in the verb, in
additon to which the ergative marker is omitted. In Ainu, the Agent has become an
optional argument in (664) as a result of the passivization of the verb. The only
indicator of ‘transitivity rearranging’ in this case is the promotion of the Patient to
the subject slot. Examples from Sre also illustrate a passive, but the given examples
differ from (663) and (664) in that the function expressed also makes a contribution
to the expression of arguments. The morphological changes in the verb in (666)
have made it possible to omit the Agent altogether. As can be seen, the expression
of Agent is optional. However, example (666) can be interpreted as both passive and
anticausative. The optional expression of Agent is possible only in the former case.
Hence, not only the verb morphology, but also the related functions condition the
expression of arguments.
Since the events in (650)-(666) possibly involve two participants, it follows
that both of them can be explicitly referred to. Moreover, this is not blocked by verb
morphology. At the extreme end of the continuum, we find alternations in which the
verb morphology makes the expression of certain argument either obligatory or
ungrammatical, cf.
 273

(667) sa-nóóta-máá-rà
3SG-knock.down-PERF-INANO
‘S/he has knocked it down’
(668) sa-nóóta-y-maa
3SG-knock.down-y-PERF
‘She has fallen down’ (Yagua, Payne 1985: 27)

(669) no-manga=mo (te pandola)

3R-eat=PF (CORE eggplant)
‘They ate (eggplants)’
(670) no-pa-manga te ika di anabou na mansuana
3R-CAUS-eat CORE fish OBL child NOM parent
‘The parent fed fish to the child’
(671) no-pa-manga te ika na mansuana
3R-CAUS-eat CORE fish NOM parent
‘The parent fed fish’
(672) no-pa-manga di anabou na mansuana
3R-CAUS-eat OBL child NOM parent
‘The parent fed the child’
(673) *no-pa-manga na mansuana
3R-CAUS-eat NOM parent
(*The parent fed) (Tukang Besi, Donohue 1998b: 3f)

In (667) from Yagua, the verb is not marked explicitly for transitivity and it takes
two arguments. In (668), the agent is eliminated from the verb semantics by
attaching the affix -y- to the verb, which intransitivizes the construction. Examples
from Tukang Besi illustrate an obligatory increase in the number of arguments as
a result of causativization. In its basic form, the verb manga allows the Patient to be
eliminated, as in (669). The number of obligatory arguments is consequently one.
The causativization of the verb increases the number of obligatory arguments by
one. One ‘non-nominative argument’ can be omitted freely, but the other has to be
expressed explicitly. Only the number is important here, since either of these two
arguments can be eliminated.
In (667) and (668), the omission of arguments corresponds to the omission of
participants. Because the given changes are explicitly marked on the verb, argument
expression is not structurally optional as, for example, in (653) and (657).
274

Semantically (or ontologically), these two alternation types coincide, but the crucial
difference is that in the latter cases the expression of arguments is verb-determined.
We might perhaps say that cases like (668) illustrate the most typical examples of
these kinds of case. This is due to the underlying semantic motivation. In addition
to the semantically motivated cases, it is also possible that the sole number of
arguments and not core participants is conditioned by the morphology of the verb.
Examples are provided below, cf.

(674) ta’wach wiich-i-m tœka-qa-‘u

man.SUBJ knife-OBJ-INSTR eat-ANT-he
‘The man ate with a knife’
(675) wiich-i-m tœka-ta-qa-ax
knife-OBJ-INSTR eat-PASS-ANT-it
‘Someone ate with a knife’ (Agent may not be expressed) (Ute, Givón 2001: 131)

(676) o 8u-l b cx koše

that.OBL-ERG grass.ABS mow.PRES
‘S/he mows the grass’
(677) cg koše-laa (*b cx-o-d)
that mow-ANTIP (*grass-OBL-INSTR)
‘She is mowing (*the grass)’ (Hunzib, Van den Berg 1995: 110)

In both cases, the number of participants is two. However, the expression of Agent
(Ute) or the Patient (Hunzib) is blocked by the verb morphology. The syntactic
(in)transitivity of these two cases is motivated differently. Clauses like (668) are
intransitive, since the event involves one participant and one possible referent only.
On the other hand, arguments are eliminated solely morphosyntactically instead of
being omitted from the cognitive structure of the event in (675) and (677). The
intransitivity is consequently motivated at different levels. In the case of genuine
intransitivizations, clauses are intransitivized at the level of semantics, which
naturally makes the explicit reference to either participant impossible. In languages
like Ute and Hunzib, on the other hand, the marking of the verb completely
intransitivizes the clause at the level of morphosyntax. Thus, there is also a
transitive construction available, but in these particular cases the expression of one
argument only is possible. For the purpose of distinguishing these constructions
 275

from each other, we label the latter as de-transitive constructions (see 5.3.1.1.2.).
Constructions given above illustrate the most complete degree of de-transitivization,
since arguments are obligatorily omitted.
In (667)-(677), the number of arguments is directly conditioned by the verb
morphology. Depending on the construction, this might coincide with a change in
the number of participants. However, not only the morphology of the verb
contributes to the expression of arguments, but the functions expressed also
contribute to whether arguments are expressed or not. There are cases in which
arguments can be expressed, even if this seemingly contradicts the morphological
marking of the verb. This is closely related to the expression of multiple functions
by a single mechanism. The following examples from Lango and Amharic (see also
(666) from Sre) exemplify cases in which a derived ‘intransitive’ verb behaves
differently depending on the nature of the related function, cf.

(678) àtîn òcègò dóggólâ

child 3SG.close door
‘The child closed the door’
(679) dóggólâ òcègérê
door 3SG.close.MID.PERF
‘The door closed’
(680) gwôk òkààyê (kénê)
dog 3S.bite.MID.PERF (self.3SA)
‘The dog bit itself’ (Lango, Noonan 1992: 132f)

(681) b cr-u (b c-t’vn vk’k’ak’e) (b c-lij-u) tc-k cffctc

door-DEF (with-care/attention) (by-boy-DEF) ANTIC/PASS-open.PERF.3M
‘The door opened/was opened (with care) (by the boy)’79
(682) aster t-at’t’cb c-
Aster REFL-wash.PERF-3F
‘Aster washed herself’
(683) lcmma ras-u-n tc-la ’ ’c
Lemma head-POSS.3M.ACC REFL-shave.PERF.3M
‘Lemma shaved his head’ (*Lemma shaved himself)

79
Amberber does not include the Agent in this particular example, but uses it in others. W e have
included the agent here for convenience.
276

(684) lcmma ras-u-n m ctta

Lemma self-POSS.3M-ACC hit.PERF.3M
‘Lemma hit himself’
(685) s cww-o -u vrsb crs-a cw(*-in) tc-d cbadd cb-u
person-PL-DEF each.other-POSS.3PL(*-ACC) RECIP-hit.RECIP.PERF-3PL
‘The people hit each other’ (Amharic, Amberber 2000: 315, 325ff)

Examples (678) and (679) from Lango illustrate a typical anticausative alternation
(glossed as middle voice by Noonan). Since the agent has been eliminated from the
cognitive structure of the event, the clause is intransitive. In (680), the verb appears
in the same morphological form, but the alternation is not anticausative, but
reflexive. Even if reflexives involve one participant only, there are still two semantic
roles and two possible referents for arguments present. As a result, the (optional)
expression of a reflexive pronoun is possible in (680). The examples from Amharic
illustrate the rather general nature of the prefix tc-. It can express a variety of
functions that (among other things) differ from each other in the number of possible
referents.80 Example (681) illustrates the difference between the anticausative and
passive readings. If the reading is anticausative, both elements (i.e the adverb
bct’vnvk’k’ak’e and the Agent bcliju) are excluded due the complete absence of the
agent. However, the ‘intransitivity’ of the verb morphology is restricted to function,
since the agent can be expressed (yet in an oblique case form), if it is a part of the
given event. Also ‘agentive adverbials’ like bct’vnvk’k’ak’e introduce an agent
implicitly. Examples (682) and (683), which illustrate different reflexives, are very
interesting as regards the correlation of verb morphology with the expression of
arguments. Clause (682) is completely intransitive. Similarly to (681), the verb form
per se does not exclude the expression of other arguments, but also reflexively
interpreted clauses can have Patients in the accusative, as in (683). Example (683)
differs from (680) from Lango, because (683) does not exemplify a typical reflexive.
This clause is grammatical only in the reading given in single brackets, i.e. only in
a ‘restricted reflexive sense’ (our own label) in which only a part of the patient is
affected (in this case the head). Amharic also has a transitive reflexive in which the
accusative Patient rasun functions as a reflexive pronoun ((684)). As a result, we

80
This is used as an umbrella term for participants and semantic roles.
 277

may claim that in Amharic the mere reflexivity is not a sufficient criterion for
excluding the expression of Patients, but the nature of the reflexive has to be taken
account of as well. Example (685) illustrates a reciprocal construction in which the
expression of Patient is possible, but in which the marking deviates from that in
typical transitive clauses. This is very natural, since the verb appears in a less
transitive form.

5.2.5. Alternations affecting the number of participants, number of arguments and

individual transitivity features

Above, we have discussed and illustrated different kinds of alternation primarily

from a structural viewpoint. Alternations were divided into transitivity increasing,
transitivity decreasing and transitivity rearranging alternations. In principle,
transitivity increase or decrease is completely neutral as regards whether the number
of arguments corresponds to changes in the number of participants.81 In similar vein,
the alternations discussed in this section are neutral as regards whether they increase
or decrease the transitivity of clauses. The semantic differences in the motivation
have already been discussed (in passing) in the previous section. In this section, we
will briefly illustrate some relevant differences between the three alternation types
listed above. Since so many examples of all the types have been illustrated already,
we will focus on the theoretical basis of the alternations below. Furthermore, the
analysis is rather cursory, since the nature of different alternations is discussed in
more detail later.
The most typical examples of alternations that impinge on the number of
participants are causatives, anticausatives, resultatives and reflexives. In the first
case, the number of participants increases, whereas the three other alternations
reduce the number of participants. Causatives differ from the other alternations
under study, since causativization is not restricted to one clause type only, but many
different clauses can be causativized (see, e.g. (491)-(496) from Bote). The number
of arguments can thus increase from one to two, two to three etc. The other
alternations differ from each other in how the decrease is motivated and also the

81
Even if the conditioning principle is different in different cases.
278

semantic roles of added participants are different. In the first two, the eliminated
participant is agent, while in reflexives the remaining participant can be regarded as
an agent.
Alternations affecting the number of participants can be divided into two
based on whether the participants in question are regarded as core or not. In the most
typical cases, participants involved belong to the core. The alternations usually have
direct consequences for the nature of events, i.e. the degree of transitivity is affected.
For example, the transitivity of the event ‘someone dies’ is crucially different
depending on whether the event in question occurs spontaneously or has been
externally instigated. The resulting state is the same. The other alternation affecting
the number of participants is exemplified by cases in which participants added or
eliminated are peripheral. The semantic peripherality of participants usually
correlates with the degree of semantic effect the alternation has on the event in
question. In many languages, this is also reflected in the morphosyntax of arguments
(e.g. in many languages the syntactic status of arguments expressing location is
lower than that of those referring to causers). The most peripheral participants in this
respect are circumstantial adverbs expressing, for example, instrument, location or
time. Applicatives in (480)-(486) exemplify this alternation type.
The participants added to the clause core as a result of applicativization do not
have any consequences for the semantic transitivity of events, even if the structural
transitivity (or rather valency) increases. The participants added are semantically
peripheral and usually refer to circumstants that are irrelevant for transitivity.
Consequently, these kinds of alternation illustrate a kind of hybrid type in which the
valency of clauses increases without any major effect on semantic transitivity.
Because of the obligatory valency increase, it would perhaps be better to label these
as alternations that primarily affect the number of arguments (see below for a more
detailed analysis). Their structural consequences are far more obvious than the
semantic ones. This dual nature leads us to discussing the next type of transitivity
alternation, namely alternations that impinge on the number of arguments only.
What is important here is that the number of (core) participants remains constant,
even if the number of arguments is affected. The most typical examples of these
alternations are perhaps provided by passives and antipassives that omit the Agent
or the Patient from (the core) of the clause without omitting them from the cognitive
 279

structure of events. In most radical cases, the morphosyntactic omission is complete

(cf. (674)-(677) from Ute and Hunzib). More typically, however, the omitted
arguments can be expressed, yet usually in oblique forms. Other kinds of case are
discussed in what follows.
Alternations that affect the number of participants and/or the number of
arguments (only) are very different in status. All languages have some means of
expressing the former kind of alternation. This has to be the case, since the
structural changes coincide with differences in the ontology of events. In the
simplest cases, the only difference between the clauses is the number of arguments
that coincides with the number of participants (cf. e.g. (595) and (596) from
Lezgian). Even if the alternation is not marked explicitly on the verb, we are dealing
with a transitivity alternation, because the number of participants changes.82 On the
other hand, languages diverge crucially in whether the omission or introduction of
arguments brings about a genuine transitivity alternation. A very simple, yet
illustrative example of these differences is illustrated below, cf.

(686) lautanen rikko-utu-i

plate.NOM break-ANTIC-3SG.PAST
‘The plate broke’
(687) hän rikko-i lautase-n
s/he.NOM break.TR-3SG.PAST plate-ACC
‘S/he broke the plate’
(688) lautanen rikko-utu-i puutarha-ssa
plate.NOM break-ANTIC-3SG.PAST garden-INESS
‘The plate broke in the garden’
(689) *puutarha rikko-i lautase-n
garden.NOM break.TR-3SG.PAST plate-ACC
(The garden broke the plate) (Finnish)

(690) no-hoko-mate-‘e na kene-su mohoo iso

3R-FACT-dead-3OBJ NOM friend-1SG.POSS sick yon
‘He killed my sick friend (permanently)’

82
It might be notoriously difficult to state whether we are dealing with a transitivity increasing or
decreasing alternation, though.
280

(691) no-mate-ako te buti

3R-die-APPL CORE fall
‘They died in a fall’ (Tukang Besi, Donohue 1999: 225, 206)

In Finnish, the difference between inchoative and causative versions of the event
‘break’ is expressed by a change in the number of arguments which is accompanied
by changes in the verb morphology. The introduced Agent receives core status. The
situation is very different in (688). In this case, the added adjunct refers to a
peripheral participant not related to semantic transitivity. In Finnish, the semantic
peripherality correlates with a lower structural status. The adjunct added in (688)
cannot be promoted to a core status, as the ungrammaticality of (689) shows.
Finnish is far from being an exceptional language in this respect. In this and similar
cases, the expression of adjuncts is completely optional (both semantically and
structurally). In Tukang Besi, on the other hand, not only causativization ((690)), but
also the introduction of other kinds of argument (pure adjuncts in other languages)
can result in a structural transitivization, as shown in (691). Finnish illustrates a
language in which the transitivization of clauses usually corresponds to a semantic
transitivization, whereas Tukang Besi exemplifies the type in which structural
transitivization is independent of semantic transitivization.
Languages with morphological applicatives can be divided into groups based
on the degree of obligatoriness and the nature of the argument resulting in an
applicativization (semantic transitivization usually coincides with structural one and
we ignore this aspect in the following discussion). First, languages can be divided
into two on the basis of whether the applicativization is the only means of
introducing certain peripheral arguments to clauses or not (see also Peterson 1998:
41ff). The first type is illustrated e.g. by Hoava (see (644) and (645)). In Hoava, the
applicative construction is the only way of permitting particular semantic arguments
to appear in clauses (Davis 1997: 237). The other type is illustrated below, cf.

(692) sa-duu rá-viimú

3SG-blow INAN-into
‘He blows into it’
(693) sa-duu-tá-ra
3SG-blow-APPL-INAN:OBJ
 281

‘He blows it’ (Yagua, Payne 1997: 187)

(694) saya mem-bawa surat itu (kepada ali)

I TRANS-bring letter the (to Ali)
‘I brought the letter (to Ali)’
(695) saya mem-bawa-kan ali surat itu
I TRANS-bring-BEN.APPL Ali letter the
‘I brought the letter to Ali’ (Indonesian, Peterson 1998: 42, cited from Chung 1976: 41)

Peterson (1998: 41) labels examples above ‘optional applicative constructions’. The
primary function of the applicativization is not to introduce arguments, but rather
to promote peripheral arguments to the core. Peripheral arguments can be introduced
without applicativization in both Yagua and Indonesian.
Languages that have some kind of applicative constructions (defined very
loosely) can further be divided into two based on the status of arguments that results
in a transitivization. More typical examples are illustrated by cases in which the
arguments receive core status.83 For example, (691) illustrates this type. The
structural transitivity increases, since the number of core arguments is affected. The
other type is illustrated by cases in which the adding of certain structurally
peripheral arguments affects the verb morphology. An example is illustrated below,
cf.

(696) kafé āngei§-na-rua ni

coffee sell-DISTPST-1SG CERT
‘I sold coffee’
(697) kafé pablo-se § k-āngei§-na-rua ni
coffee Pablo-LOC PERI-sell-DISTPST-1SG CERT
‘I sold coffee to Pablo’ (Ika, Frank 1990: 69)

According to Frank (ibidem), the introduction of the beneficiary is obligatorily

accompanied by what he has labelled as ‘peripheral participant valence increase’
(glossed PERI). This is not a typical applicative, since the introduced argument does

83
It is in this case irrelevant, whether the applicativization of the verb follows from the core status of
the added argument or whether the applicativization is necessary for adding arguments to the clause.
282

not receive core status. These kinds of case present a kind of intermediate form
between or a combination of true applicatives and free introduction of arguments,
like (601) and (602). They have ‘inherited’ the verbal marking from applicatives and
the form of arguments from languages lacking applicatives. This kind of applicative
seems to be attested far more rarely than the typical ones. Examples above illustrate
the only instance attested in our data.
Above, we have examined different kinds of alternation affecting the number
of participants and/or arguments primarily from the perspective of verb morphology
and its correlation with the obligatoriness of the expression. A further very
important aspect of whether we can label a change in the number of arguments as
a transitivity alternation is illustrated by the morphological marking of arguments.
As shown in (688) from Finnish, the lower morphosyntactic status of arguments is
also reflected in the argument marking. The introduced arguments are marked in a
core case only in genuine transitivizations. In these kinds of case, the status of the
argument enables us to label a given construction as a transitivity alternation. There
is no circularity involved, if we assume that high semantic transitivity coincides
with high linguistic transitivity. However, the core vs. non-core marking of
arguments is far from being an unequivocal transitivity criterion. First, it is
inapplicable to languages that lack case marking altogether, because of which all
arguments are morphosyntactically identical and we cannot distinguish between true
and ‘false’ transitivizations. Second, even languages with morphological cases allow
some non-core participants to appear in core cases, cf.

(698) mies rikko-i ikkuna-n

man.NOM break-3SG.PAST window-ACC
‘The man broke the window’
(699) mies juoks-i tunni-n
man.NOM run-3SG.PAST hour-ACC
‘The man ran for an hour’ (Finnish)

(700) ifá-t wo "hk-ís

dog-NOM bark.LGR-IND
‘A dog is barking’
(701) ifá-t pó "si lást-i"-n á "ssi"c-ís
dog-NOM cat black-DUR-OBL chase.LGR-IND
 283

‘A dog is chasing a black cat’

(702) ifá lást-i"-t fítta-n hôy |-is
dog black-DUR-NOM outside-OBL stand.FGR-IND
‘A black dog is standing outside’ (Creek, Martin 2000: 378)

In Finnish, certain adverbials expressing time (and also measure or amount) can
appear in the accusative as well. Example (698) illustrates a genuine transitivity
alternation. In this case, an agent is added to an originally intransitive (inchoative)
event, which transitivizes the underlying intransitive clause. Structurally (699) also
exemplifies a transitivity alternation, even (699) cannot be regarded as one
semantically. In Creek, most non-subject arguments appear in the oblique case
regardless of the semantic roles of participants. It is thus impossible to distinguish
between true and ‘false’ transitivizing alternations structurally. Only semantics
provides us with sufficient means to do this. Excluding alternations like (699) and
(702) seems intuitively justified, since we are clearly dealing with an introduction
of participants that do not contribute to transitivity of events.
Examples (699) and (702) illustrate cases in which an argument is added to
the clause, which does not, however, correspond to a change in the number of core
participants. Further examples of this are the external possession and body part
ascension constructions illustrated in (624)-(629). Typical passives and antipassives
are also examples of transitivity alternations that primarily affect the number of
arguments. The exact opposites of these alternations are illustrated by alternations
that affect the number of participants without having any consequences for the
number of core arguments. In these cases, the number of core participants either
increases or decreases, which does not, however, affect the number of arguments.
Examples of this were already illustrated in light of transitivity rearranging
alternations illustrated in (636)-(641). A couple of further examples are given
below, cf.

(703) as kni:ga yecna jieša n

I.ERG book.NOM bought read.INF
‘I bought a book to read’
(704) as kni:ga yecna jieši:ta n
I.ERG book.NOM bought read.CAUS.INF
284

‘I bought a book for someone else to read’ (Chechen, Nichols 1994: 68)

(705) hã' tapé thá-de-bo

1SG.NOM rice eat-TENSE-CAUS 1
‘I (causee) was made to eat rice (by someone – unstated but implied causer)
(706) hã' tapé thá-syig-a
1SG.NOM rice eat-CAUS 2-TENSE
‘I (causer) made (someone – unstated but implied causee) eat rice’ (Mishmi, Dixon 2000: 47,
cited from Sastry 1984: 155f)

(707) mies syö-tt-i liha-n laps-i-lle

man.NOM eat-CAUS-3SG.PAST meat-ACC child-PL-ALL
‘The man fed the meat to the children’
(708) mies syö-täty-tt-i liha-n laps-i-lle
man.NOM eat-CAUS-CAUS-3SG.PAST meat-ACC child-PL-ALL
(??/* jaako-lla)
( Jacob-ADESS)
‘The man made someone (Jacob) feed the meat to the children’
(709) mies syö-täty-tä-tt-i liha-n laps-i-lle
man.NOM eat-CAUS-CAUS-CAUS-3SG.PAST meat-ACC child-PL-ALL
(??/* jaako-lla)
( Jacob-ADESS)
‘The man made Jacob made someone feed the meat to the children’ (Finnish)

(710) a-b c a-r Ø-g’a-y-c-8aa pw-a p-s

3-OBL 3-ABS 3-HOR-3-NONPRES-see-PAST-AFF
‘He saw him’ (HOR=horizon of interest)
(711) a-b c yar c-r z-y-c-8aa pwc-ž-a p-s
3-OBL self-ABS REFL-3-NONPRES-see-self-PAST-AFF
‘He saw himself’ (Kabardian, Colarusso 1992: 137)

Examples from Chechen, Mishmi and Finnish illustrate causatives derived from
transitive clauses. In Chechen, the number of participants increases from two to
three which is, however, not reflected in the number of arguments which remains
two. The causee is implicit in (704). In Mishmi, there are two different causatives
neither of which allows both the causer and the causee to be expressed explicitly,
but either is (depending on the causative marker) only implicitly referred to. In case
we wish to express both the verb has to be stated twice. Example (707) from Finnish
 285

also illustrates a typical causative derived from a transitive clause. In this case, the
number of participants correlates with the number of arguments. Example (708)
illustrates a double and (709) a triple causative. Morphological causativization of
the verb always adds a participant to the event denoted. If the number of participants
exceeds three, the explicit referring to each participant becomes less felicitous. It is
only marginally grammatical to refer to all four or five participants in these cases
and usually only three of them are stated explicitly (cf. Kibrik 1996: 133 for
Godoberi). Consequently, at least one participant is only implicitly involved. This
is very likely due to the notoriously difficult processing of these examples.
Examples (710) and (711) illustrate a transitive reflexive of Kabardian. As noted on
many occasions above, reflexives are intransitive as regards the number of distinct
participants, even if they involve two possible referents. This dual nature is directly
reflected in the typology of reflexives. There are languages with intransitive
reflexives and languages that express reflexivity transitively. In Kabardian, the
Patient is substituted by the pronoun ‘self’. Hence, the Kabardian reflexive is very
similar to that of English. The case frame remains transitive in (711).
The third type of transitivity alternation to be discussed here is represented by
alternations that affect only some individual transitivity features. These kinds of
alternation are here distinguished from the two types illustrated so far, since they do
not affect the number of arguments or participants. The consequences for transitivity
of clauses and events are consequently usually less drastic. Typical features of this
kind are, for example, agency and the affectedness of the patient. It is also possible
that the alternations have consequences for both of these features, cf.

(712) sta=re-s shing(*-la) ‘chad-pa

axe-ERG tree cut
‘Cut the tree with an axe’
(713) shing-la sta=re-s gzhus-pa
tree-LOC axe-ERG hit
‘Hit the tree with an axe’ (Lhasa Tibetan, examples courtesy of Scott DeLancey)

(714) di 'o besun bito

1SG.OBL knife.NOM lost
‘I lost the knife (not my fault)’
286

(715) di besun bitalo

1SG.ERG knife.NOM lost.CAUS
‘I lost the knife (my fault)’ (Tindi, Kibrik 1985: 307)

(716) õangula-yu algi- õ a `i wagu:ray

he-ERG kill-PAST this.ABS kangaroo.ABS
‘He killed this kangaroo’
(717) a `i bu `u:r wayi-`i daba:y-a
this boy.ABS afraid-PRES dog-LOC
‘The boy is afraid of the dog’ (Yaygir, Crowley 1979: 374f)

The examples from Lhasa Tibetan illustrate the distinction between ‘hit’- and
‘break’-verbs (see Fillmore 1970). In Lhasa Tibetan, typical Patients appear in the
absolutive, whereas those referring to participants not necessarily affected at all
appear in the locative case, as in (713). This difference in marking coincides with
a difference in the affectedness of the patient. Examples from Tindi illustrate a
mirror image of (712)-(713), since the affected parameter is the agency, whereas the
alternation has no consequences for the affectedness parameter. Examples from
Yaygir illustrate the type in which both parameters are affected.
Examples (712)-(717) illustrate what might be labelled as ‘iconic
alternations’. The marking appears on the argument that refers to the participant
affected by the alternation. For example, in Lhasa Tibetan, the parameter impinged
on is the affectedness of the patient and the only difference between the clauses is
the marking of Patient. The iconicity is naturally not an inherent feature of the
alternations, but there are also many non-iconic ones. Examples include the
following, cf.

(718) ngaa-§i waa § saay§-naa-ng

I-ERG bird.ABS hear-PRES-1SG
‘I hear a bird’ (non-agentive)
(719) ngaa-§i waa §-kaay § saay§-naa-ng
I-ERG bird-DAT hear-PRES-1SG
‘I listen to a bird’ (agentive) (Chepang, DeLancey 1981: 634)

(720) õatu yinana danka-na wara-yi

1SG.ERG 2SG.ABS find-PART AUX-PRES
 287

‘I have found you (purposefully)’

(721) õani danka-tadi-na wara-yi yiõka õu
1SG.ABS find-ANTIP-PART AUX-PRES 2SG.LOC
‘I have found you accidentally’ (Diyari, Austin 1981: 154)

(722) mies tappo-i karhu-n

man.NOM kill.PAST-3SG bear-ACC
‘The man killed the bear’
(723) isä rakast-aa jaakko-a
father.NOM love-3SG.PRES Jacob-PART
‘Father loves Jacob’ (Finnish)

In (718) and (719), the affected transitivity feature is agency. As DeLancey points
out, the former example is non-agentive, while in the latter the degree of agency is
higher. Since a change in the agency parameter is signalled on the Patient, we are
dealing with a non-iconic alternation. Examples from Diyari and Finnish are also
non-iconic, but in different ways. The change in Diyari is more or less similar to that
illustrated in (718) and (719). In Chepang, the non-iconicity is due to the flagging
of the alternation on the ‘wrong’ argument. In Diyari, on the other hand, the
alternation is ‘more than ideally’ marked, since the marking of both arguments
changes, even if only one of them is semantically affected. In (722) and (723), both
transitivity features are affected. The Agent (i.e. the nominatively marked argument)
refers in (722) to a typical agent and the Patient to a semantic patient. In (723), on
the other hand, the Agent refers to an experiencer and the Patient to a stimulus. The
non-iconicity refers here to the fact that a change that affects both participants is
marked solely on one of them. Hence, the given change is ‘less than ideally’
marked.
The existence of non-iconic alternations can perhaps be explained by referring
to the structure of languages. Also the motivation of marking makes a contribution.
In Diyari and Finnish, this is quite obvious. In (720) and (721), the non-iconicity is
likely due to the antipassive verb morphology. Antipassivization requires that both
arguments be marked differently from those in transitive clauses. However, only the
agency parameter is affected, which results in the ‘redundant’ marking. In Finnish,
the (nominative) subject relation is quite strong. One result of this is that only one
288

of the arguments in transitive or ditransitive clauses has to appear in the nominative.

Consequently, it is irrelevant whether one or more participants are affected as a
result of an alternation, since the marking of only of them can be manipulated.
Because of this non-iconic cases like (723) above cannot be avoided.
The three alternation types illustrated above affect the transitivity of clauses
at different levels. Alternations that affect the number of participants occur at the
ontological level, because of which the participants in question are either integral
parts of events or they are completely eliminated from them. Changes in the number
of participants are accompanied by obvious changes in the number of transitivity
features as well. Alternations that omit participants typically result in a complete
omission of all the relevant transitivity features associated with the eliminated
participant. Conversely, if a participant is introduced to an event at the ontological
level, the relevant features are usually introduced, too. For example, in the case of
anticausative and causative alternations, the agency features are either completely
eliminated from an event (anticausative) or they are added to it as a whole
(causative). Alternations that affect the number of arguments (only) do not affect the
number of participants. Cases in which changes in the number of argument coincide
with changes in the number of participants are here (because of the primacy of
semantics) regarded as instances of participant adding/omitting alternations.
Alternations that affect the number of arguments occur at the level of morphosyntax,
which means that they cannot have any effect on the number of participants nor do
they have consequences for any ontological aspect associated with participants.
Typical examples are illustrated by passives and antipassives that decrease the
number of arguments. Despite the complete omission of an argument, the
corresponding participant remains an integral part of the event denoted. For
example, the omission of the Agent does not necessarily decrease the degree of
agency. We may even say that omitted Agents are usually interpreted as typical. The
case is naturally different, if the passive marker that results in the omission forces
a different interpretation (cf. (521) from Bahasa Indonesia). These alternations
should perhaps be divided into two depending on whether the changes are
accompanied by morphological changes in the verb. This is important, since the
marking vs. non-marking of alternations correlates in some cases with the
obligatoriness vs. optionality of the expression or omission of arguments. In the case
 289

of the mere introduction or omission of arguments without any morphological

marking on the verb, the expression of arguments is usually completely optional (in
case it can in theory be optional). This means that, for example, in English clauses
he eats the meat, he eats and he eats (meat) in a restaurant are equally possible. The
actual expression of arguments occurs first at the level of morphosyntax. On the
other hand, if the omission or introduction of an argument results in morphological
changes in the verb (or if this enables the expression or omission), the expression
of arguments is more directly conditioned by the predicate. Either the morphological
changes result in changes in the possibly expressed arguments (as in passives or
antipassives) or the verb morphology makes the expression of arguments impossible
(cf. Ute and Hunzib above) or obligatory (many applicatives). Since the third
alternation type is neutral for the number of participants or arguments, they can
occur at both levels. Typical examples of alternations at the ontological level are,
for example, alternations that change the inherent semantics of events. In these
cases, the morphosyntactic expression of events varies according to ontological
differences. Obvious examples of alternations that occur at the level of
morphosyntax are alternations that are due to changes in definiteness of arguments
(cf, e.g. (375)-(376) from Turkish).
The occurrence of alternations at different levels also makes the different
alternations completely independent of each other. Put concretely, this means, for
example, that both causative and passive alternations can be applied to one and the
same clause. This does not result in any misinterpretations, since the consequences
for the structure of clauses and events are very different and apply at different
phases of derivation. Passivization does not ‘neutralize’ the effect of causativization,
since passivization only affects the number of arguments. A couple of examples that
illustrate this independence are given below, cf.

(724) §a-thi§
3SG.S-die
‘He died’
(725) §a-ka-thi§-hno §
3SG.S-1SG.O-die-MAL
‘He died on me’
290

(726) §an-vaa-lee khaa §an-§ii-tha §-hno §-hnaa

3PL.POSS-husband-PL DEIC 3PL.S-REFL-kill-MAL-PL.OBJ
‘They killed their husbands to the detriment of each other’ (Haka Lai, Peterson 1998: 28)

(727) š-Ø-kam le: ts'i§

ASP-3SG.ABS-die the dog
‘The dog died’
(728) š-e:-qa-kam-isa:-x
ASP-3PL.ABS-1PL.ERG-die-CAUS-TR
‘W e killed them’
(729) š-Ø-kam-isa-tax le: utiw q-uma:l
ASP-3SG.ABS-die-CAUS-CM PL.PASS the coyote 1PL.POSS-by
‘The coyote got killed by us’ (K’iche’, Campbell 2000: 239, 277, 248)

(730) t-c-walom-p§a-n kawr ctk-c-n

1SG-E-hear-TH-3SG footstep-E-3SG.ABS
‘I heard the footsteps’
(731) pcmnan t-c-n-walom-at- c-nat cnp cnacp-c-t
1SG.ERG 1SGA-E-CAUS-hear-TH-E-3PL.O old.man-E-3PL.ABS
‘I informed the old men (i.e. caused the old men to hear/understand)’
(732) pcmo t-ena-n-walom-at- c-k
1SG.ABS 1SG-ANTIP-CAUS-hear-TH-E-1SG
‘I made an announcement’ (Chukchi, Dunn 1999: 212, 216)

(733) wane e laba-tani-a maka nau

man 3SG.NFUT harm-TRANS-3SG.OBJ father 1SG
‘The man harmed my father’
(734) roo wane kero kwai-laba-taqi
two man 3DU:NFUT DEPAT-harm/spoil/etc-TRANS
‘The two men harm (people), spoil, damage (things) etc.’ (Toqabaqita, examples courtesy of
Frank Lichtenberk)

Examples from Haka Lai illustrate two kinds of applicative construction. The first
of these is derived from an intransitive clause. As can be seen, the introduction of
peripheral participants as core arguments at a ‘non-ontological level’ does not affect
the transitivity of the event itself. Consequently, the lexeme employed is intransitive
in (725). On the other hand, in (726) the lexeme is transitive, since the event
denoted involves two core participants. The additional argument refers in both cases
 291

to a maleficiary. Examples (727)-(729) illustrate the independence of causative and

passive alternations. In K’iche’, causativization is marked by the affix -isa:. This
affix appears also in (729) that illustrates a passive derived from (728). Examples
from Chukchi and Toqabaqita exemplify a case in which an antipassive is derived
from a clause explicitly marked as transitive. The antipassivization does not
intransitivize the verb morphology.
Even if the threefold division illustrated above may seem rather clear, there
are also many less obvious cases in addition to numerous clear ones. For example,
typical causatives are obvious examples of alternations that affect the number of
participants. On the other hand, passives like (675) from Ute and antipassives like
(677) from Hunzib are clear examples of alternations that only affect the number of
arguments, whereas the likes of (712)-(717) illustrate alternations that only affect
some individual transitivity features. Besides, there are cases in which it is
extremely difficult to state explicitly what kind of alternation we are dealing with.
These difficulties were briefly discussed above, where it was noted that some
applicatives are an intermediate form between alternations that affect the number
of participants and alternations that merely introduce arguments. They do add
participants, but these are not core and do not contribute to the transitivity of the
given events in any significant way. On the other hand, morphological applicatives
do increase the number of core arguments. Above, we adopted the view that
applicatives are best analyzed as argument adding alternations, since their structural
consequences are far more obvious than their effect on semantic transitivity.
However, there are also applicatives that make a more significant contribution to the
overall transitivity of events. Examples of this include the following, cf.

(735) gurrungga burra-ng

kookoburra fly-PRES
‘The kookaburra flies’
(736) gurrungga-nggu djama burra-rri-l
kookoburra-ERG snake fly-TRANS-PRES
‘The kookaburra flies off with a snake’ (Djabugay, Patz 1991: 297)

(737) rul=ni§ ka-§in= §a § §a-ka-lu §-hno §

snake=ERG 1SG.POSS-house=LOC 3SG.SUBJ-1SG.OBJ-enter-MAL
292

‘A snake came into my house on me’ (Lai, Peterson 1998: 99)

The introduced arguments refer above to participants that do contribute to

transitivity of events. In both (736) and (737), the introduced argument refers to a
participant somehow affected by the profiled event. In (736) the location of the
given participant changes. In (737), the introduced participants is an indirectly
affected maleficiary. The two languages exemplified above differ from each other
in the nature of the marker employed in applicativization. In Djabugay, the
applicative affix can also causativize clauses (see Patz:ibidem). In (736), we are
clearly dealing with an applicative construction, since the S of (735) coincides with
the A of (736). In Lai, there are 7 morphologically distinct applicative suffixes (cf.
(480)-(486)) of which hno§ introduces a maleficiary. If applicative markers are
morphologically distinct, we can distinguish between these different cases by
referring to semantics. For example, in case the added argument expresses location,
it does not qualify as an alternation that introduces core participants. On the other
hand, if the added argument is an affected participant, as in (737), the alternation
can more justly be labelled as genuinely transitivizing. It may seem unjustified or
ad hoc that different instances of an alternation type are classified differently
depending on their semantic consequences. The distinction can, however, be
considered justified in languages like Lai in which we can exclude the non-typical
cases based on morphological criteria. Here we a problem less. This does not,
however, enable us to conclude what kinds of participant should be regarded as
having sufficiently significant consequences for transitivity to be viewed as core. In
the present study, the viewpoint chosen is semantic despite the obvious
controversies. This is due to the existence of languages like Lai in which we can
distinguish participant adding alternations from purely argument adding ones on a
morphological basis. This is in line with the exclusion of ‘transitivizations’ like
(702) in which all (or most) introduced arguments are core regardless of their
semantic roles. Non-morphologically distinct applicative constructions (e.g.
Kichaga, see Peterson 1998: 39 and Hoava, see Davis 1997: 236ff) are here regarded
primarily as argument adding alternations that can occasionally increase the number
of core participants.
Another problematic case is illustrated by direct vs. indirect causatives. In
 293

prototypical cases, the causativization results in a complete transitivization of events

(and clauses). In this case, an archetype of agent is introduced to an intransitive
event whose only participant is a patient. There are also causatives that do not
transitivize events completely, since the agent is not prototypical or the causation
is in some other way less typical. What makes these constructions as possibly
problematic for the threefold division proposed in the present section is that they can
be regarded either as two independent causatives or the less transitive variant can
be viewed as a subtype of the causative archetype, in which case we are rather
dealing with an alternation that affects individual transitivity features. Examples are
illustrated and discussed below, cf.

(738) ’ikay y-exu-s

glass.ABS II-break-PAST.W IT
‘The glass broke’
(739) už-~ ’ikay y-exu-r-si
boy-ERG glass.ABS II-break-CAUS-PAST.W IT
‘The boy broke the glass’
(740) uži-q ’ikay y-exu-s
boy-POSS glass.ABS II-break-PAST.W IT
‘The boy accidentally broke the glass’ (Tsez, Comrie 2000: 365)

(741) tráak háan

buffalo die
‘The buffalo died’
(742) rwàay p-háan tráak
tiger CAUS-die buffalo
‘The tiger killed the buffalo’
(743) k c̀c tòk háan múuc
3SG.M CAUS 2 die ant
‘He happened to kill an ant’(Kammu, Svantesson 1983: 103f)

(744) nuqa fan-ta rumi-ta apa- i-ni

1SGA Juan-ACC rock-ACC carry-CAUS-1SGA
‘I made Juan carry the rock’
(745) nuqa fan-wan rumi-ta apa- i-ni
1SGA Juan-INSTR rock-ACC carry-CAUS-1SGA
‘I had Juan carry the rock’ (where Juan voluntarily submits to the causer’s wishes) (Quechua,
294

Dixon 2000: 66, cited from Cole 1983: 118)

In Tsez, there are two causativizing mechanisms. In the typical case ((739)), a
causative affix is attached to the verb, which transitivizes the construction in
question. The Agent appears in the ergative and the Patient in the absolutive. In
(740), the causativization is less complete both semantically and structurally. The
verb is not causativized and the Agent does not appear in the ergative, but in the
possessive case. In Kammu, the two causatives are distinguished on the basis of the
form of the causative affix or particle. Prototypical causation is designated by the
causative prefix p-, as in (742). Less direct (or accidental) causation, on the other
hand, is expressed periphrastically, as in (743). This corresponds to the typological
generalization that direct causation is usually morphological and less direct
causation periphrastic (see e.g. Comrie 1989: 172f). In (738)-(743), it is not
appropriate to claim that the less direct causatives are derived from prototypical
ones, since the structural differences are so obvious. Hence, it is rather safe to state
that in languages like Tsez and Kammu there are two individual causativization
mechanisms (and consequently two different transitivizing alternations) that both
add an instigator to an intransitive event, but that diverge in the degree of agency
associated with the agent. Examples from Quechua also illustrate two different
causatives. In this case, it is more obvious that we are dealing with two instances of
one general type. The verb affix is the same and the only difference between the two
constructions is found in the marking of the causee. The differences attested in (744)
and (745) are due to changes in the degree of volitionality associated with the
causee. Hence, these are very obvious examples of alternations that affect individual
transitivity features instead on being two different instances of alternations that
affect the number of participants. As opposed to Tsez and Kammu, this analysis is
supported by structural evidence. Furthermore, differences in transitivity are
expressed similarly to differences in individual aspects (i.e. by merely manipulating
the marking of arguments).
Examples (740) and (743) illustrate cases in which an increase in the number
of participants does not fully transitivize events. Moreover, there are cases in which
an additional participant does not increase the semantic transitivity, but can even
reduce the overall transitivity. Applicatives that add peripheral circumstants are the
 295

most obvious examples of cases in which increases in the number of participants are
completely insensitive to the transitivity of clauses. These are not discussed any
further here, since a more detailed analysis is found above. The emphasis below lies
on cases in which the introductions of participants reduces the overall transitivity
of events, the purpose of which is also to illustrate the independence of valency and
semantic transitivity. The most typical examples of this are provided by causatives
of transitive clauses. Since causativization introduces an agent or causer to the
event, the (semantic) valency of verbs always increases. Prototypically, the
causativization of intransitive events coincides with a complete transitivization (cf.,
however, (738)-(745) above). This is not the case, if an agent (or a causer) is added
to a transitive or a ditransitive clause, cf.

(746) ašak ool-du etteen

old.man.NOM boy-ACC hit.PAST
‘The old man hit the boy’
(747) bajyr ašak-ka ool-du ette-t-ken
PN old.man-DAT boy-ACC hit-CAUS-PAST
‘Bajyr made the old man hit the boy’ (Tuvan, Sumbatova 1993: 254)

(748) m ã-d wad ’eme baläx

father-ERG mow.AOR grass.NOM
‘The father has mowed the grass’
(749) m ã-d wat ’emnäwne gezal-s baläx
father-ERG mow.CAUS.AOR son-DAT grass.NOM
‘The father made the son mow the grass’ (Svan, Sumbatova 1993: 258)

(750) Çla lep p hnanak ximd’

child bread his.older.sister give.FIN
‘The child gave his older sister the bread’
(751) ctck Çla-ax lep p hnanak ximgud’
father child-CAUSEE bread his.older.sister give.CAUS.FIN
‘The father made/let the child give the bread to his older sister’ (Nivkh, Nedjalkov et al 1995:
78)

In (746)-(751), the number of participants increases either from two to three or from
three to four. Differently from causatives of intransitive events, this does not
296

correlate with an increase in the semantic transitivity. If we define high transitivity

as an interaction of two distinct participants, one of which volitionally instigates the
event which impinges on the other (as we do in the present study), alternations like
(746)-(751) rather degrade the basic transitivity of (monotransitive) events.
Examples (746), (748) and (750) illustrate typical transitive or ditransitive events
instigated by volitionally acting agents. In Tuvan, the Agent of transitive clauses
appear in the nominative case and in Svan, in the ergative, whereas in Nivkh the
Agent (as well as other arguments) are morphologically zero. In (747), (749) and
(751), the marking of the original Agent, which refers to the participant responsible
for carrying out the action profiled, changes. In Tuvan, the Agent appears in the
dative instead of the nominative in (747). In Svan, the change is from ergative to
dative, while in Nivkh the causee appears in a special form used solely for causee
marking (cf. Comrie 1976: 274).
The changes in the marking of Agent noted above may be claimed to reflect
changes in the semantic role of the agent (cf. below). What seems to be important
for causativization of transitive events is a kind of ‘hierarchy of command’. This
means that one of the participants involved in the instigation of events ranks higher
than other(s). Put concretely, this means, for example that in (749) the father has the
authority to tell his son to mow the grass. If the son does what is ordered, he submits
himself to the authority of the father (i.e. the causer). If both participants involved
in the instigation equally agentive, both are very likely marked as Agents, in which
case the verb is not causativized. For example, the clause he made me paint the
house profiles a hierarchically organized instigation, while he and I painted the
house is more appropriate, if the participants are equally ranked for agency (cf. also
the relevance of asymmetry for transitivity in general). This hierarchy is typically
reflected somehow in the marking of causer and causee. The causer usually occupies
the primary argument slot and the original Agent is demoted in status (cf. Comrie
1976: 263) and it appears, e.g. in the dative (as in Svan and Tuvan), instrumental
(Hungarian), locative (Daghestanian languages), allative (West Greenlandic),
adessive (Finnish) or possessive (Tsez) (this list along with the language names is
taken from Dixon 2000: 55). This distribution of marking reflects the hierarchy
noted. The causer outranks the causee and it follows that the causer is assigned the
highest grammatical status. On the other hand, since the marking is motivated by a
 297

hierarchy of command, it cannot reflect the degree of involvement in the actual

action. The causee is clearly more involved in the transitive action in cases like
(747), (749) and (751), which makes it more similar to the agent of typical transitive
events. Hence, it would do more justice to the causee to be marked identically to the
transitive Agent (even if the agency is also semantically reduced in these cases, cf.
(1087)-(1094) below). This is cross-linguistically rare, though, since the marking is
conditioned by other principles. In some rare cases, though, the causer and the
causee do bear the same marking, cf.

(752) alaweru-k hai-ts axos disi ka

Alaweru-ERG 1SG-ERG child.ABS beat CAUS
‘Alaweru made me beat the child’ (Trumai, Guirardello 1999: 353)

In Trumai, the original Agent has retained the ergative marking despite the
introduction of a causer. The ergative marking of the causer reflects the ‘hierarchy
of command’, whereas the ergative marking of the causee is due to the features
shared with a typical agent. Structurally, (752) resembles a co-ordinated non-
causative noted above. Consequently, it would be tempting to interpret (752) as
having two agents both of which are carrying out the same action and having the
exact same status without one outranking the other.
As noted above, the causee is usually distinguished from the transitive agent
only through a lower degree of volitionality. The will of the causee is manipulated,
and a complete volitionality is lacking. Since the causer ranks highest in the
hierarchy, it acts completely volitionally, which also implies that it has the greatest
potentiality of preventing the event from happening. The degree of volitionality of
causee can vary from mildly affected cases, in which the causee is acting
volitionally, but in which the decision to act comes ‘from above’, to a very strong
(perhaps even physical) manipulation. The strength or directness of the causation
is inversely propositional to the degree of volitionality associated with the causee
(cf. Premper 1988: 55). The stronger the manipulation on the causee is, the less
volitionally the causee acts. As shown in (744) and (745), these differences can also
be reflected in the marking of causees. In other respects, the causee retains its
agentive status (in case it is a typical agent to start with). It is in control of its own
298

actions and it intends the action to be targeted at a patient (i.e. it is not accidentally
causing the event to occur). However, as a result of introducing an additional agent,
the original agent loses certain agency parameters to the causer, which produces a
less transitive event.
The hierarchical organization of instigation also implies that the two
participants involved in instigating and carrying out an action are responsible for
different facets. As noted above, the introduced agent is most directly responsible
for the initiation of events that eventually result in a change-of-state in the patient.
The causee, for its part, carries out the action ordered ‘from above’. In causatives
derived from transitive clauses, the relevant features related to transitive agents are
divided (unequally) between two distinct participants. The causation deprives the
causee of complete volitionality which becomes a property of the causer. This
different distribution implies a certain ‘agency limit’ that cannot be exceeded. In
causatives derived from intransitive clauses, the agency is lacking altogether,
because of which the introduced agent is typical. Hence, the volitionality has shifted
from causee to causer, and it no longer is a feature of the causee. The causee retains
all the other relevant agency features, since they are integral parts of agency and
typical transitive events in general. Further evidence for the existence of an agency
limit is provided by the use of non-causatives, if neither participant outranks the
other in agency. In these cases, the two agents are conceived of as a single agent or
the event is conceptualized as two different events involving a single agent each. We
are not dealing with any kind of ‘double agency’ in these cases.
The kind of limit illustrated above in light of agency can be applied to the
affectedness parameter as well. Similarly to a certain amount of agency features,
transitive events also have one directly affected patient only. This is not say that, the
number of affected participants can be only one, but that there can be only one entity
or a group of entities directly affected. This means that we do not exclude events
like ‘he ate three pieces of bread’ from our definition. Events like these are here
regarded as single transitive events, even if they consist of three ‘subevents’ (of
‘eating’). Hence, other participants can be only indirectly affected by events. They
do not constitute the primary targets in events they are affected by. Examples
include beneficiaries or maleficiaries both of which have been illustrated above.
Unlike causatives of transitive events, the introduction of indirectly affected
 299

participants does not have any consequences for the nature or degree of the
affectedness parameter. This is also reflected in the marking of Patients. In
distinction from Agents, Patients usually retain their marking and grammatical
status despite the introduction of further arguments. In this respect, alternations that
introduce indirectly affected participants to events can more justly be regarded as
genuine transitivizing alternations structurally. The introduced arguments usually
rank lower for their grammatical status. A couple of somewhat different examples
of this are given below, cf.

(753) hän raken-si talo-n (minu-lle)

3SG.NOM build-3SG.PAST house-ACC (1SG-ALL)
‘S/he built a house for me’ (Finnish)

(754) ne-ni (towe-pe) §ufiya fei- §afe-a-§a

3SG-ERG (Towe-BEN) sweet.potato boil-PROG-3SG-IND
‘She’s cooking sweet potatoes for Towe’ (Tauya, MacDonald 1990: 121)

(755) cá "ni-t istaha "kocí-n ha "y-ís

John-NOM doll-OBL make:LGR-INDIC
‘John is making a doll’
(756) cá "ni-t cími-n istaha "kocí-n ín-ha "y-ís
John-NOM Jim-OBL doll-OBL DAT.APPL-make:LGR-INDIC
‘John is making a doll for Jim’ (Creek, Martin 2000: 390)

Examples above illustrate the two basic ways of encoding arguments that refer to
indirectly affected participants. In the first case, exemplified by Finnish and Tauya
(cf. also (463) and (464) from Japanese), the Patient and the added beneficiary or
maleficiary bear different marking. In Finnish, both arguments are marked explicitly
for case. The Patient appears in the accusative, whereas the beneficiary appears in
the allative case.84 In Tauya, the Patient has zero (absolutive) marking and the
beneficiary argument appears in a special benefactive case in (754). Creek illustrates
a language type in which Patients and arguments referring to indirectly affected
participants are identical in form. As examined above, the most important function

84
This is the case of indirect objects also in inherently ditransitive clauses.
300

of applicatives is to enable the introduction of core arguments referring to peripheral

participants. Thus, constructions like (756) are rather frequent in languages with
morphological applicatives. It is, however, noteworthy that the Patient usually
retains many or most of its monotransitive properties. The most important of these
for our purposes is the case marking that is only rarely affected as a result of
applicativization. This means that the basic transitivity of the original clause is
retained and not decreased as is the case in most causatives. There are, however,
also cases in which the Patients lose some relevant properties to the added
argument. An example of this is the following, cf.

(757) xiv-ãn ja-’ui’-ca-‘ gu-tatracui’

now-1SG.S 3PL.DO-go to(PL)-APPL-FUT ART-chickens
‘I’ll take the chickens away right now’
(758) xiv-ãn jum-’ui’-dy-ica-’ gu-tatracui’
now-1SG.S 2SG.DO-go to(PL)-APPL-FUT ART-chickens
‘I’ll bring the chickens to you (SG) right now’
(759) xiv-ãn jam-b v-idy-ica-’ gu-tacárui’
now-1SG.S 2PL.DO-go to-APPL-FUT ART-chicken
‘I’ll bring the chickens to you (PL) right now’ (Tepehuan, Peterson 1998:51, cited from W illet
1981: 68)

As illustrated above, the verb agrees in terms of person and number with the Patient
in monotransitives and with the beneficiary/recipient in (758) and (759). Hence, the
grammatical status of the Patient has decreased in the latter two examples. Instead,
the beneficiary is treated in the same way as the typical Patient in (757). If we
assume that the verb agreement correlates with affectedness, the indirectly affected
participant is considered directly affected in (758) and (759).
As illustrated above, Agents and Patients of monotransitives behave in
radically different ways as regards the consequences of introducing additional agents
or patients. The status of the original Agent is much more frequently than not
affected by the introduction of a causer. On the other hand, Patients usually retain
their case marking, even if their status may be affected in other ways (as in (758)
and (759), for example). This might, at least to some extent, be explained by
referring to the more frequent occurrence of distinct causative constructions in
 301

comparison with morphological applicatives. However, this seems unsatisfactory,

since many applicatives do not demote Patients of monotransitives in the way
Tepehuan does. Consequently, we propose a semantic explanation. As noted above,
the introduction of a causer to a transitive event deprives the original agent of
complete volitionality, which results in a lower degree of overall agency (and
transitivity of the event). Agency is a multi-layered notion that can be divided into
relevant subcomponents including volitionality, intentionality and control. Each of
these can be lacking without this having any major consequences for the others.
Consequently, it is possible for the causer to deprive the agent of volitionality
without this resulting in a complete lack of agency. The agent is still an agent in
light of the other relevant features (e.g. it still controls the event). On the other hand,
affectedness of the patient is not as multi-layered or as readily dividable into
different components. The relevant features of affectedness include the manner and
degree of affectedness. Closely related to both is the saliency parameter, i.e. whether
we can directly observe whether the patient has been affected or not. Furthermore,
the number of relevant features does not directly coincide with the overall
affectedness, while the degree of agency is more readily computable from the
number of distinct features. If the number is three, the agent is usually more of an
agent than an entity that only has two of the relevant features. On the other hand, the
features of affectedness are always present, but their nature and intensity vary. It is
not possible to be a patient without a certain manner of affectedness. Furthermore,
what is central for the ‘fate’ of the original Agents and Patients of monotransitive
constructions is that features of affectedness are not ‘transferrable’. This means that
the introduction of patients to events cannot deprive original patients of
affectedness. In case someone gets indirectly affected as a result of someone getting
hit, the ‘sympathy’ does not impinge on the direct affectedness. In the world we live
in, it is not possible to think of a situation in which, for example, the direct effect
of breaking is divided between two participants one of which is responsible for the
degree of affectedness and the other for the manner. This is not to say that
affectedness is a constant notion. As shown in many examples throughout the
present study, it may vary. What is important, though, for our purposes is that the
unrealized part of affectedness is not transferred to another participant. Since the
introduction of patients does not have any effect on the semantic nature of the
302

original patient, it is rather consequent that changes illustrated in (758) and (759) are
less frequent cross-linguistically. Furthermore, they are usually restricted to changes
in grammatical status of the Patient (and not the patient). As regards the transitivity
of events, the true, directly affected patient makes a more significant contribution
to the overall transitivity (along with the nature) of events than the indirectly
affected patients. They are crucial for the occurrence and nature of transitive events
in general. For example, the nature of breaking follows from the fact that something
gets broken. It is irrelevant whether someone gets indirectly affected by this or not.
Hence, directly affected patients outrank indirectly affected ones in the ‘patient
hierarchy’. In similar vein, causers are more responsible for the occurrence of
transitive causatives.

5.2.6. A structural typology of transitivity alternations

Above, we have discussed transitivity alternations from different perspectives. The

two most important divisions presented were the two threefold divisions into
transitivity increasing, transitivity decreasing and transitivity rearranging
alternations on one hand and to alternations that increase or decrease the number of
participants or arguments or affect individual transitivity features on the other. The
former is primarily based on the structural consequences and the latter on the
underlying semantic motivation of the alternations. Despite discussing transitivity
alternations from various perspectives, we have not proposed a detailed structural
typology of alternations yet. This is the goal of the present section in which we try
to illustrate (as exhaustively as possible) the ways in which the structural transitivity
of clauses can be manipulated. The typology proposed is a detailed description of
transitivity increasing, transitivity decreasing and transitivity rearranging
alternations. The relevant structural features include the number and marking of
arguments along with the verb morphology. The constituent order is ignored in the
sense that an illustrated instance of a type does not have to display the very order
illustrated in order to be included in the type. (e.g. the order can be Patient-Agent,
even if the description suggests A O V). For simplicity, the order in the descriptions
is always Agent-Patient. However, the order is regarded as a way of distinguishing
Agent and Patient in languages that lack case marking (see e.g. 1a). Furthermore,
 303

as shown by Hiirikoski (see Hiirikoski 1991, 1992 and 2002), there are cases in
which word order signals changes in transitivity and is sensitive to it. It is, however,
quite unlikely that this is the only, or even primary transitivity indicator in any
language. This means that we do not expect to come across a language in which
SVO-clauses are highly transitive, whereas OVS-clauses are not, and that lacks all
other morphosyntactic ways of distinguishing highly transitive from less transitive
constructions. Also in Finnish (studied by Hiirikoski), other mechanisms (case
marking and verb morphology) are more important in this respect. Word order is
merely an optional mechanism of transitivity expression. In general, we may
conclude that the more transitive a clause is the harder it inverts into OVS-order (see
Hiirikoski 1992: 192). The illustration of the different types below is loyal to the
threefold divisions illustrated above. Analyses concerning the existence of different
types are limited to a minimum, even if every alternation is also briefly discussed
in addition to the mere illustration. Alternations that affect the number of originally
ditransitive clauses will not be regarded as an independent type (the outcome may,
however, be ditransitive). This is primarily due to their similarities with alternations
affecting clauses with two arguments, because of which studying them in detail
would not make a significant contribution to our typology. We begin our illustration
by transitivity increasing alternations. Alternations affecting intransitive clauses are
discussed before those affecting transitive clauses.
Below the notation goes as follows. The symbols used in the 9 main types
refer to arguments of intransitive, transitive and (derived) ditransitive clauses
without any reference to a specific role. For example, in type 1 X may refer both to
Agent and Patient85 depending on the basic marking pattern of a given language. X
and Y refer loosely to Agent and Patient (but not necessarily in this order), whereas
the third term Z covers all other arguments. If there is only one argument present,
it is X, the first added argument is Y and the second Z. It is, however, in order to
stress that these symbols are not associated with semantic roles, but the semantic

85
Even if noted above, it may be appropriate to stress here that ‘Agent’ and ‘Patient’ refer here
primarily to grammatical roles, whereas ‘agent’ and ‘patient’ are used exclusively for semantic roles.
Below, Agent refers to any argument that signals the instigator of an event (that does not have to be
a typical agent) and Patient refers to any core other core participant (that can deviate from the
prototype of a patient), as well.
304

role of an agent may be referred to by Z as well. We have opted for using different
symbols for the main and specific types for the purpose of minimizing the
misinterpretations due to the misuse of the given symbols. Symbols S, A, O and E
are used much in accordance with Dixon and Aikhenvald (cf., e.g. 2000: 3) with
some modifications. Hence, S refers to the sole argument of an intransitive clause,
A to the Agent, O to the Patient and E to any other argument. Structurally, different
instances of S, A, O and E are not distinguished. This means, for example, that
nominatively and ergatively marked Agents are equal in their status as Agents. In
their basic use, symbols S, A, O and E refer to grammatical functions (in the case
of A and O the semantic role associated with the arguments is also of the utmost
importance). This means that S refers to the function of intransitive subject (in a
loose sense) in basic clauses. Symbol A refers to the Agent in a transitive clause
irrespective of the marking of the argument in question (cf. above). In similar vein,
O is the argument that refers to the patient in the event denoted. E refers to any
argument that is marked differently from A and O. Also in this case, the exact
marking is irrelevant, which means that we do not distinguish partitive, dative,
locative etc. arguments from each other, but they are all referred to as E.
The mere symbols S, A, O and E without any further modification in subscript
refer to cases in which the form and the function coincide. The highly language-
specific typical form of arguments is based on the signalling of the functions in
basic clauses. This means that the typical form of S is inferred from the marking of
the sole argument in intransitive clauses. If the clause, in which the argument
appears in, is intransitive, we are dealing with a typical S. In similar vein, symbols
A and O are based on the linguistic reference to the agent and patient of transitive
events in basic transitive clauses. The agent is referred to by A and the patient by O.
What is of the utmost importance, however, is that the semantic roles of agent and
patient may be referred to in many different ways and the mere use of A and O does
not always suffice. Symbol E refers to the marking of any argument that is
structurally different from S, A and O. The functions vary greatly (E can also refer
to agent and patient), but in basic cases the function cannot be an agent or a patient.
The following simple illustration exemplifies the basic use of the symbols, cf.
 305

the man came

S V
the man hit him for me
A V O E

Restricting the use of the symbols to the basic sense does not suffice to describe the
vast variety of structurally different transitivity alternations attested cross-
linguistically. Thus, we have widened the scope. S (without any further
modifications, see below) refers only to the sole argument of clauses denoting
events that have only one core participant. This means that the ‘subject’ of passives
is not a mere S in the present context. In similar vein, the use of A and O requires
that the event denoted have two core participants (both of which do not have to be
explicitly referred to). For example, in distinction from Dixon and Aikhenvald the
argument of he eats is an A instead of an S, since it refers to an agent in the profiled
event and, more importantly, it is structurally identical to the Agent in he eats the
meat. One of the goals of this unorthodox use is to show that, at the level of specific
alternations, accusative and ergative languages display clear differences.
In addition to using symbols S, A, O and E in a more semantic way than
Dixon and Aikhenvald, the symbols are used in subscript as well (as can be seen on
numerous occasions in the typology below). This follows from the inadequacies of
the basic use. Subscript is used, if the form and the semantic role referred to
disagree.86 In these cases, symbols in normal script refer to the form and those in
subscript signal the semantic role referred to. A concrete example should clarify
this. In the English passive, the O is promoted and it appears in a zero marked
nominative case. Consequently, the argument at issue is similar to both A and S.
The A argument is demoted to E status. It seems intuitively inadequate to refer to
a zero marked patient as A87 due to the semantic ‘burden’ of the label. The semantic
role of the agent is still present (even if the argument is omitted) and it would be odd
to use the symbol A for the patient function. Thus, the semantically neutral S is
favoured in all cases in which we have to choose between two possible terms. The

86
It has to be emphasized anew here that S in the basic use is neutral in this respect.

87
There are also cases in which there clearly are two explicitly marked A’s present, cf. below.
306

passive of English is illustrated and the symbols discussed below, cf.

he was hit by the man

SO V-TR EA

Differently from the typical S of a semantically intransitive clause, the primary

argument of passive is marked SO. The label means that the given argument is
formally identical to the typical S, but the semantic role referred to is different from
the typical one. The marking EA means that an oblique argument refers to the
semantic role of agent. This is not typical of E in English, since the reference is
typically to non-core participants like benefactives and location. Consequently, it is
important to explicitly flag this difference by adding the corresponding symbol in
subscript. It merits a note that in the typology below, E is explicated by a symbol in
subscript only if the role referred to by it is a typical agent or a patient. In all other
cases, E is not modified any further.
In what follows, different alternation types will be illustrated using the
notation explained above. A couple of further remarks on the verb morphology are
also in order. Verb morphology includes also verbal cross-reference in most cases,
since it is usually a ‘by-product’ of changes in the number or status of arguments
and can consequently be ignored as an independent feature. As regards labelling the
verb morphology as transitive (+TR) or intransitive (-TR), it is irrelevant whether
the (in)transitivity is explicit or implicit. Also cases in which the transitivity is
signalled by omitting a(n) (in)transitivity marker are also taken account of.

The typology

1. X V ÷ X Y V(+TR)

This type includes all the alternations that result in a complete transitivization of an
originally intransitive clause. Even if the shift from intransitive to transitive
constructions can justly be regarded as the most typical instance of transitivization,
this alternation type will be discussed only briefly. This is due to the very detailed
typology of basic transitive clause proposed above. Hence, most of the relevant
 307

aspects of (in)transitivization in this respect have already been discussed. In

distinction from the other illustrated types, different alternations are not subdivided
depending on whether the verb morphology has been affected or not. All the
examples presented below illustrate the typical transitive construction of the
languages under study. We can distinguish five different subtypes on the basis of
argument marking in transitive clauses. In the first four cases, the symbol in
boldface refers to the argument(s) that is (are) marked differently from S. In the first
type, i.e. (1a), both arguments of the transitive clause appear in the same
morphological form as S and their order is the only way of distinguishing between
them. In (1b), the Agent of transitive clauses is structurally identical to S, and O is
marked differently (usually in the accusative case). In (1c), the Patient has retained
the marking of S, while the Agent appears in a non-zero case form. In (1d), neither
of the arguments in transitive clauses has zero marking. (1e) comprises all the
possible cases in which there are two morphologically distinct transitivizing
mechanisms available used depending on the related function. The differences in
transitivization do not result in differences in argument marking.

1a. S V ÷ A O V (+TR)

(760) tráak háan

buffalo die
‘The buffalo died’
(761) rwàay p-háan tráak
tiger CAUS-die buffalo
‘The tiger killed the buffalo’ (Kammu, Svantesson 1983: 103f) 88

1b. S V ÷ A O V (+TR)

(762) der mann starb

ART.NOM man die.PAST.3SG
‘The man died’

88
As regards the argument marking pattern, this type has to be regarded as NOM-ACC, since S and
A both precede the verb.
308

(763) der bär tötete den mann

ART.NOM bear kill.PAST.3SG ART.ACC man
‘The bear killed the man’ (German)

(764) aannan-ni daanf-e

milk-NOM boil-AGR
‘The milk boiled’
(765) terfaa-n aannan daanf-isa-a
Terfa-NOM milk boil-CAUS-AGR
‘Terfa boiled the milk’ (Oromo, W atters 2000: 213)

As can be seen, this type illustrates typical nominative-accusative marking pattern.

Examples from Oromo are illustrated here in order to show that accusative can also
be the unmarked case form.

1c. S V ÷ A O V (+TR)

(766) ga `i `ilmu ira

tree break-PAST
‘A tree broke’
(767) mawun-du ga `i `ilmu man-i
man-ERG tree break-PAST
‘A man broke a tree’ (Djaru, Tsunoda 1987: 192)

This is the mirror image of 1b, i.e. the pattern in absolutive-ergative. Differently
from obvious marked nominatives, marked absolutives do not seem to exist (cf.,
however, Corston 1996: 15).

1d. S V ÷ A O V(+TR)

(768) tarppa-Ø pak-tu-wa-n

tree-NOM break-AFS-AFV
‘Tree broke’
(769) yuul= õu-Tu tarppa-Na pak-tu-ma-n
man-ERG tree-ACC break-AFS-CAUS-AFV
‘Man broke tree’ (Yuulngu, Schebeck 1976: 382)
 309

1e. S V ÷ A O V +TR1/TR2

(770) s v kuo:rta loz

my head.NOM hurts
‘My head aches’ (Ingush, Nichols 1982: 446)
(771) cuo sv kuorta loza-bu
he.ERG my head(B).NOM hurt-(B)make
‘He’s hurting my head’ (e.g. by hitting it, direct causation)
(772) cuo sv kuorta laza-boaqq
he.ERG my head(B).NOM hurt-(B)take
‘He’s always doing things to make my head hurt’ (less direct causation, Ingush, Nichols
1994b: 118, 124)

(773) bake-Ø-ra ani-ai

child-ABS-AS become.big-INC
‘The child is growing’
(774) e-n-ra bake-Ø ani-a-i/ani-ak-ai
1-ERG-AS child-ABS become.big-TRNZ-INC
‘I raise the child’
(775) e-n-ra bake-Ø ani-ma-ai
1-ERG-AS child-ABS become.big-CAUS-INC
‘I make the child grow’ (Shipibo-Conibo, Valenzuela 1997: 108)

Type 1e covers here all possible cases in which different verb morphology
distinguishes between two instances of the same construction type. In order to save
space, we have not distinguished the four types presented in 1a-1d here. Typical
examples of 1e include direct and less direct causatives.

2. X V ÷ XZV

In the alternations illustrated thus far, the transitivization is complete, and the result
is a typical transitive structure. In the other possible type that introduces arguments
to intransitive clauses, the transitivization is less complete, since the introduced
argument appears in a form untypical of the argument at issue in basic clauses.89

89
This comprises all cases apart from nominative, absolutive, accusative (and/or dative) and ergative.
310

There are five subtypes illustrated below, cf.

2a. S V ÷ SOV

(776) mawun (daru) man-an

man.ABS (Djaru) talk-PRES
‘The man talks (Djaru)’ (Djaru, Tsunoda 1995: 97)

In this type, both arguments appear in a zero marked case form. The type is therefore
seemingly similar to 1a from Kammu. However, what makes it justified (and
necessary) to regard these two types as different is that (776) does not illustrate the
typical transitive clause of Djaru (cf. (767)), while (761) exemplifies the typical
transitive pattern of Kammu. The marking of Agent differs from the typical ergative
Agent in (776). In 2a, S is used non-typically, since it is not marked for a semantic
role by a symbol in subscript. This is due to the fact that the argument might also
refer to a patient in this kind of structure (in cases like ‘the tree fell on the house’).

2b. S V ÷ SA E V(+TR)

(777) mastov mak qet-in vai-n

enemy PVB attack-AOR 1PL-DAT
‘The enemy attacked us’ (Batsbi, Holisky & Gagua 1994: 193)

(778) mare ätpiral (“lile-s”)

man.NOM he.sing.II (“Lile”-DAT)
‘The man sang (“Lile”)’ (Svan, Harris 1985: 131)

Type 2b illustrates what Dixon & Aikhenvald (e.g. 2000: 3) label as ‘extended
intransitive’. Both languages illustrated below are predominantly ergative as regards
the marking of arguments. However, in (777) and (778) the Agent does not appear
in the ergative, but has retained its marking despite the introduction of a new
argument. In addition, the E does not refer to a typical patient, which is reflected
directly in the marking.
 311

2c. S V ÷ A E V(+TR)

(779) ¨ali-di rasul- ’u (tata) tudi

Ali-ERG Rasul-CONTESS (saliva) spit.PAST
‘Ali spat (saliva) at Rasul’ (Godoberi, Kibrik 1996: 144)

(780) yarvin y-omrok

woman 3SG.DISTPST-steal
‘the woman stole’

(781) yarvin y-omrok õi namas

woman 3SG.DISTPST-steal OBL clothes
‘The woman stole the clothes’ (Ura, Crowley 1998: 28f)

Differently from 2b, the Agent is structurally a typical Agent in 2c. What makes
these constructions less transitive is the marking of the Patient.

2d. S V ÷ E O V(+TR)

(782) get’e xa-na

pot break-AOR
‘The pot broke’
(783) zamara.di-waj get’e xa-na
Zamira-ADEL pot break-AOR
‘Zamira broke the pot accidentally/involuntarily’ (Lezgian, Haspelmath 1993:292)

2d has a typical O, but the Agent appears in the adelative instead of the typical
ergative.

2e. S V ÷ E SO V(+TR)

(784) kuppi puto-si

cup.NOM drop-3SG.PAST
‘The cup dropped’
(785) minu-lta puto-si kuppi
1SG-ABL drop-3SG.PAST cup.NOM
312

‘I accidentally dropped the cup’ (Finnish)

Type 2e differs from 2d only in that the marking of Patient is different from the
typical O. The Agent appears in the ablative instead of the nominative. Hence, the
Patient cannot appear in the accusative. Differences between 2d and 2e are largely
due to differences in the basic argument pattern of the given languages.

3. X Y V ÷ X Y Z V(+TR)

Types 1 and 2 illustrate transitivity alternations that increase the number of

originally intransitive clauses. Type 3 comprises transitivized monotransitives. The
most typical alternations of this kind are provided by causatives and applicatives.
As noted and discussed above, these alternation types diverge in the nature of the
introduced argument. This also manifests itself in the consequences the introduction
has for the clause structure. In the illustration below, the arguments inside the single
square brackets refer to the arguments in the original clause (the order Agent-Patient
is retained).

3a. A O V ÷ [A O] OE V

(786) nganaju yaan (pawulu-ngara-a) kampa-lalha murla-a

1SG.GEN wife (child-PL-ACC) cook-PAST meat-ACC
‘My wife cooked meat for the kids’ (Martuthunira, Dench 1995: 68)

(787) aringkila iltja awiya utnhika

dog.ERG hand.NOM boy.NOM bite.PAST
‘The dog bit the boy’s hand’ (Alyawarra, Yallop 1977: 119)

The only difference between the monotransitive and the derived ditransitive clause
here is the existence of an additional O argument that refers to a peripheral
participant. Both arguments in the original clause have retained their marking and
status.
 313

3aa. A O V ÷ [A O] OE V+TR

(788) cá "ni-t istaha "kocí-n ha "y-ís

John-NOM doll-OBL make:LGR-INDIC
‘John is making a doll’
(789) cá "ni-t cími-n istaha "kocí-n ín-ha "y-ís
John-NOM Jim-OBL doll-OBL DAT.APPL-make:LGR-INDIC
‘John is making a doll for Jim’ (Creek, Martin 2000: 390)

(790) kì-bid-ítsé bàná

SUBJ.1SG-call-ANT CL2.children
‘I have called the children’
(791) kì-bíl-éd-ítsé bàná dìj ]
SUBJ.1SG-call-APPL-ANT CL2.children CL8.meal
‘I have called the children for meal’ (Tswana, Creissels 2000: 237)

As opposed to 3a, the argument introduction is here accompanied by morphological

changes in the verb. The two exemplified languages differ from each other in the
marking of the two Patient-like arguments. In Creek, they are both explicitly marked
for case, while in Tswana they are morphologically unmarked and are not cross-
referenced on the verb.

3aaa. A O V ÷ AE [OA O] V+TR

(792) nuqa fan-ta rumi-ta apa- i-ni

1SGA Juan-ACC rock-ACC carry-CAUS-1SGA
‘I made Juan carry the rock’ (Quechua, Dixon 2000: 66, cited from Cole 1983: 118)

Type 3aaa is here classified as a subtype of 3a on the basis of the identical outcome.
As in 3a and 3aa, the resulting structure is A O O V. What distinguishes 3aaa from
3a and 3aa is the semantic role of the introduced argument, which also affects the
marking of the original arguments. In 3aaa, the first O refers to the agent in the
original event instead of a patient.

3b. A O V ÷ AE [A O] V+TR
314

(793) alaweru-k hai-ts axos disi ka

Alaweru-ERG 1SG-ERG child.ABS beat CAUS
‘Alaweru made me beat the child’ (Trumai, Guirardello 1999: 353)

In 3b (that is very marginally attested in the languages of the world), the only
difference between transitive and derived ditransitive clauses is the appearance of
an additional A. The marking of the underlying monotransitive has been retained.

3c. A O V ÷ SE [A O] V+TR

(794) antha waya-rnda untu maka ‘nthunha tyintya-rda-nha-nga

I want-PRES you.ERG firewood mine cut-SP-NONP-LOC
‘I want you to chop up my firewood’ (W angkangurru, Hercus 1994: 185)

(795) qa the:-wu peitse-e-ze ze-pe-za

1SG 3SG-AGT cup-one-CL DIRECTION-buy-CAUS.1SG
‘I made him/her buy a cup’ (Qiang, Dixon 2000: 50, examples courtesy of Randy LaPolla)

In 3c, the introduced causer appears in a morphologically unmarked form, whereas

the original Agent retains its marking. The division of Agent properties is not
perfect in (795), since the introduced causer is cross-referenced on the verb as is the
typical Agent. Thus, (795) could also be considered a subtype of 3b, since there are
two A’s. We have, however, distinguished between these based on the marking on
the independent arguments.

3d. A O V ÷ E [A O] V

(796) nabisat.a-z ruš.a ktab k’el-na k’an-zawa

Nabisat-DAT girl.ERG book read-AOC want-IMPF
‘Nabisat wants her daughter to read a book’ (Lezgian, Haspelmath 1993: 297)

In 3d, an argument in the dative (i.e. E) is added to a transitive clause. In this

particular case the E is very similar to a causer. However, differently from a typical
causer, it does not appear in the ergative (cf. Haspelmath 1993: 276), but in the
 315

dative. This difference is due to the semantics of the verb ‘want’. The Agent appears
in the dative instead of ergative in general with this verb.

3e. A O V ÷ AE [OA SO] V+TR

(797) aster s vga-w-vn k'orr ct'c-

Aster meat-DEF-ACC cut.PERF-3FEM
‘Aster cut the meat’

(798) lcmma aster-vn (s vga) as-k'orr ct'-at

Lemma Aster-ACC (meat) CAUS-cut.PERF.3MASC-3FEM.O
‘Lemma made Aster cut (some meat)’ (Amharic, Amberber 2000: 320)

The marking in (797) and (798) is in a way the exact opposite of 3c, since O is
demoted to the S status. In typical transitive clauses, O appears in the accusative, as
in (797). In Amharic, causativization usually demotes the O to S for the purpose of
avoiding double accusative clauses.90

3f. A O V ÷ [A O] E V

(799) tunu e au e ika (ma sione)

cook ERG 1SG ABS fish (for Sione)
‘I am cooking the fish for Sione’ (Niuean, Seiter 1980: 36)

(800) ich wasch-e mein-e hände

I.NOM wash.PRES-1SG 1SG.POSS-PL hands
‘I wash my hands’
(801) ich wasch-e mir die hände
I.NOM wash.PRES-1SG I.DAT ART.PL hands
‘I wash my hands’ (German)

As opposed to the previously illustrated types, the peripheral argument is a typical

E argument both structurally and functionally in 3f. Hence, E is an optional, freely

90
According to Amberber (798) would be grammatical also with two O’s, but in cases like this, the
original O is typically demoted.
316

omissible adjunct.

3ff. A O V ÷ [A O] E V+TR

(802) kafé āngei§-na-rua ni

coffee sell-DISTPST-1SG CERT
‘I sold coffee’
(803) kafé pablo-se § k-āngei§-na-rua ni
coffee Pablo-LOC PERI-sell-DISTPST-1SG CERT
‘I sold coffee to Pablo’ (Ika, Frank 1990: 69)

(804) taveti-a mosi sa lose

make.TR-3SG Mosi ART.SG room
‘Mosi built the room’
(805) tavete-ni-a kinahe mosi sa lose
make-APPL-3SG nipa.palm Mosi ART.SG room
‘Mosi built the room with nipa palm’ (Hoava, Davis 1997: 239)

Type 3ff diverges from 3f mainly in that the introduction of arguments is not
possible without a morphological modification of the verb. The two illustrated
languages differ crucially from each other as regards the status and marking of E.
Examples from Ika are rather exceptional cross-linguistically, since the introduction
of obvious E’s is accompanied by morphological changes in the verb. According to
Frank (ibidem), (803) would be ungrammatical without the verb affix k-. The
applicativization usually adds ‘core’ arguments to clauses. This is also seemingly
the case in Hoava, since all arguments appear in a morphologically zero form.
However, examples like (805) are here regarded as instances of A O E-type, since
the applied object occupies the secondary object slot in Hoava. As a result, the status
of the introduced argument is somewhat lower, even if this is not manifested
morphosyntactically.

3fff. A O V ÷ AE [OA EO] V+TR

(806) ke hara nraasi

2SG eat rice
 317

‘You eat rice’

(807) sabùrina nrâ fa-hara nrü nrî nraasi
Sabrina 3SG CAUS-eat 2SG O rice
‘Sabrina fed you rice’ (Tinrin, Osumi 1998: 115f)

(808) õw c fèe z]
he fear-PERF snake
‘He was afraid of a snake’
(809) m c fès c õw c (n c z ])
I fear-CAUS-PERF he (with snake)
‘I frightened him (with a snake)’ (Babungo, Song 1996: 173, cited from Schaub 1985: 211)

Similarly to 3aaa, examples (806)-(809) differ from those in 3f and 3ff in that the
introduced argument is a causer. As in 3aaa, differences in the nature of the added
argument result in different consequences for the semantic role assignment despite
the identical result.

3g. A O V ÷ AE [EA O] V+TR

(810) b ccca-« macho k ha-ik

child-ERG fish eat-3SG.PAST
‘The child ate the fish’
(811) ama-« b ccca-ke macho k hw-a-ik
mother-ERG child-DAT fish eat-CAUS-3SG.PAST
‘The mother fed the child fish’ (Bote, examples courtesy of Balaram Prasain)

3g illustrates perhaps the most typical causative pattern cross-linguistically. A

causer is introduced to a transitive clause, which directly affects the marking of the
original A, while the O retains its status and marking (the rationale behind this was
discussed in detail above). The introduction of the causer is typically accompanied
by causativization of the verbal complex, as also in (811).

3gg. A O V ÷ [A EO] OE V+TR

(812) naa jeefà duutsèe

1SG.PERF throw(I) stone
318

‘I threw a stone’
(813) naa jèefi birìi dà duutsèe
1SG.PERF throw(II) monkey PREP stone
‘I threw a stone at the monkey’ (Hausa, Heide 1989: 21)

(814) m-tu a-li-ki-soma ki-tabu

man SUBJ-PAST-OBJ-read book
‘The man read the book’
(815) m-tu a-li-wa-som-e-a ki-tabu wa-toto
man SUBJ-PAST-OBJ-read-APPL book children
‘The man read the book to the children’ (Swahili, Itkonen 1997: 213f)

3gg illustrates an applicative version of the most typical causative pattern. The
outcome is the same, but the introduced argument refers to a recipient or
male/benefactive instead of a causer. The change is most obvious in Hausa in which
the added (animate) target occupies the O slot and the original O is demoted to the
oblique status. The marking appears on the nominal arguments themselves. In
Swahili, the introduced arguments inherit the cross-reference from O (which is
probably due to animacy). In (814), the object agreement affix cross-references the
Patient, whereas in (815), the object affix refers to a benefactive. Hence, the
grammatical status of original O decreases. In distinction from Hausa, there is no
marking on the nominal arguments.

4. X Y V ÷ X V (-TR)

Also in type 4, the number of arguments is affected. Differently from the previous
types, transitivity is decreased in type 4. Also here, we proceed from mild to more
severe modifications of the clause structure. Cases in which the arguments are
simply eliminated without any further modifications are illustrated first and
obligatorily intransitive constructions last. Even if the main function of the
alternations illustrated here is to decrease the number of arguments, we have also
included cases in which the decrease is optional and the arguments can also be
expressed explicitly (cf. the discussion of dual nature of some alternations above).
This follows from the generally vague status of the argument omission here. Many
 319

languages allow arguments to be rather freely omitted, even if they allow them to
be expressed, if necessary. Consequently, it is in some cases extremely difficult to
conclude whether an argument is truly eliminated or not. In general, the status of the
possibly expressed arguments is, however, somewhat lower than in typical clauses.

4a. A O V ÷ A (O) V

(816) mies sö-i (leivä-n)

man.NOM eat.PAST-3SG (bread-ACC)
‘The man ate (the bread)’ (Finnish)

(817) ti t] (e kuyan t e) kupe wã

he ERG (his body DEF) wash STAT
‘He is washing (his body)’ (Shokleng, Urban 1985: 172)

In 4a, the only difference between transitive and intransitive clauses is the omission
of O. We have put the O inside brackets in order to emphasize the optionality of the
omission. Because of the differences in the basic argument pattern, the two
illustrated languages differ from each other in how explicitly A is marked for its role
as agent. In (816) from Finnish, A is structurally identical to S, whereas in Shokleng
the A argument is explicitly marked for its role.

4aa. A O V ÷ A (O) V-TR

(818) kanak=nó jangke=n pancing *(mpaq/lépang/léndóng)

child=that PRES=3 catch fish/frog/eel
‘The child is catching fish/frog/eels’
(819) kanak=nó jangke=n mancing (mpaq/lépang/léndóng)
child=that PRES=3 N.catch (fish/frog/eel)
‘The child is catching (fish/frog/eels)’ (Sasak, Austin 2000: 7)

4aa differs from 4a only in that the verb morphology has to be manipulated in order
to make the omission of O possible. In (818), the O is an obligatory part of the
clause, whereas in (819) it can be eliminated. Differently from 4a, the omission is
not completely free.
320

4aaa. A O V ÷ A V-TR

(820) a vavina i kita ra bul

ART woman she hit ART child
‘The woman hit the child’
(821) a vavina i kikita
ART woman she hit.INTR
‘The woman hit (someone or it)’ (Tolai, Mosel 1991: 248)

(822) pur-är-d u lälemx em

cow-PL-NAR PVB they.ate.it.II hay.NOM
‘The cows ate hay’
(823) pur-är-d lälemx mindor-isga
cow-PL-NAR they.ate.it.II meadow-in
‘The cows ate in the meadow’ (Svan, Harris 1985: 128f)

O remains an optional part of the clause that can be added without any further
modifications in 4a and 4aa. In 4aaa, the omission is complete and obligatory. In
both illustrated languages, the omission of O is accompanied by obvious changes
in the verb morphology. In Tolai, the verb is changed from transitive to intransitive,
and in Svan the preverb u is omitted.

4b. A O V ÷ (A) O V

(824) (matti ga) otoko o korosh-ita

(Matti NOM) man ACC kill-PAST
‘Matti/someone killed the man’ (Japanese)

(825) (ada) jak t’üna

(he.ERG) meat.ABS eat.AOR
‘He ate the meat/the meat was eaten’ (Lezgian, examples courtesy of Martin Haspelmath)

Type 4b illustrates a kind of mirror image 4a, since the omitted argument is A
instead of O. Above, A is eliminated without this resulting in any other structural
changes. Also here, we can see how the basic argument marking pattern correlates
with the obviousness of the marking. Since accusative languages mark O explicitly,
 321

it follows that they mark O explicitly also here.

4bb. A O V ÷ (EA) O V

(826) raam-nee drakht kat-yaa

Ram-ERG tree cut
‘Ram cut the tree’
(827) (raam koolõõ) drakht kat-yaa
(Ram by) tree cut
‘The tree was got by Ram’ (Panjabi, Saksena 1980: 813, 823)

In 4bb, A is demoted to E status and it can be optionally omitted. The lower status
of A distinguishes 4bb from 4b. There are no changes in the verb.

4bbb. A O V ÷ (EA) O V-TR

(828) fe’i lladdodd draig

PTCL’OBJ killed.ACT dragon
‘A dragon killed him’
(829) fe’i lladdwyd (gan ddraig)
PTCL’OBJ killed.PASS (by dragon)
‘He was killed by a dragon’ (W elsh, Comrie 1977: 55)

(830) raj-le ava-lay hirka-y-o

Raj-ERG Ava-DO/DAT hit-PAST-3SG.MASC
‘Raj hit Ava’
(831) (raj-dwara) ava-lay hirka-i-y-in
(Raj-OBL) Ava-DO/DAT hit-PASS-PAST-3SG.FEM
‘Ava was hit by Raj’ (Nepali, Givón 2001: 148)

4bbb exemplifies the other possible subtype of EA O V-type. The difference between
4bb and 4bbb is found in the verb morphology. In 4bbb, we have to intransitivize
the verb for it to be possible to omit the Agent.

4bbbb. A O V ÷ O V(-TR)
322

(832) mamach tuaci punikyay-kya

woman.SUBJ child.OBJ see-PERF
‘The woman saw the child’
(833) mamachi punikya-ta-kha
woman.OBJ see-PASS-PERF
‘Someone saw the woman/the woman was seen’ (Ute, Givón 2001: 152)

In 4bbbb, the Agent is completely eliminated. Differently from 4bbb it cannot

appear optionally in an oblique case.
It is typical of 4a and 4b that either argument is omitted without this resulting
in any other changes in the marking of the remaining argument. We proceed to cases
in which the marking of the remaining argument is also affected. Type 4c comprises
cases in which the Patient is omitted (or demoted), whereas in 4d the Agent is
eliminated (or demoted). The former illustrate different antipassives and the latter
different kinds of passive.

4c. A O V ÷ SA (O) V-TR

(834) ngaju juga-ala nyabay

1SG.ERG drink-PRES water.ABS
‘I am drinking water’
(835) ngay juga-le-la (nyabay)
1SG.NOM drink-ANTIP-PRES (water.ABS)
‘I am drinking (water) repeatedly’ (Bandjalang, Austin 1982: 38)

In distinction from the previous types, the argument marking (the marking of A)
shifts from ‘semantic’ to ‘structural’. Differently from typical antipassives, the
marking of the Patient is not affected in any way.

4cc. A O V ÷ SA (EO) V

(836) tuku-yu tuar ityayi

dog-ERG snake bite
‘The dog bites/bit the snake’
 323

(837) tuku (tuar-ku) ityayi

dog.ABS (snake-DAT) bite
‘The dog is biting (the snake)’ (Kalkatungu, Blake 1982: 86)

(838) na’e inu ‘a e kavá ‘e Sione

PRET drink ABS the kava ERG John
‘John drank the kava’
(839) na’e inu (kavá) ‘a Sione
PRET drink (kava) ABS John
‘John drank (kava)’ (Tongan, Churchward 1953: 76)

4cc illustrates a ‘non-morphological antipassive’ type (labelled as ‘antipassives

without de-transitives’ in Palmer 1994). The marking of both arguments is affected
and the Patient is an optional argument. Kalkatungu and Tongan differ from each
other in the marking of the Patient. The Patient appears in the dative in Kalkatungu,
whereas in Tongan the Patient bears no case clitic. Since the marking is different
from S, A and O, we have labelled it as E.

4ccc. A O V ÷ SA (E) V-TR

(840) hansi-p inuit tuqup-paa

Hansi-ERG people kill-IND.3SG.3SG
‘Hansi killed the people’
(841) hansi (inun-nik) tuqut-si-vuq
Hansi.ABS (people-MOD) kill-ANTIP-IND.3SG
‘Hansi killed (people)’ (W est Greenlandic, Manning 1996: 82)

(842) rìõó õ]l ùbúrr-ì õ]l-]

meat cut Ubur-ERG cut-SUF
‘Ubur will cut the meat’
(843) ùbúr õùt-ò kí rìõó
Ubur cut.CF.ANTIP-SUF PREP meat
‘Ubur will cut the meat’ (Päri, Andersen 1988: 302)

4ccc illustrates a typical antipassive alternation. The Agent is promoted to S,

whereas the marking of Patient shifts from zero to oblique. The antipassivization is
324

also signalled on the verb.

4cccc. A O V ÷ SA V-TR

(844) o 8u-l b cx koše

that.OBL-ERG grass.ABS mow.PRES
‘S/he mows the grass’
(845) cg koše-laa (*b cx-o-d)
that mow-ANTIP (*grass-OBL-INSTR)
‘She is mowing (*the grass)” (Hunzib, Van den Berg 1995: 110)

4cccc illustrates a completely de-transitive antipassive. In 4c-4ccc, an explicit

reference to the patient is possible (in a different case form), while in Hunzib the
antipassivization excludes the expression of Patient altogether.
In 4c-4cccc, the Agent is promoted to S, whereas the Patient is demoted. The
languages under study are mostly ergative languages. The Agent is explicitly marked
and most of these languages have a mechanism that promotes the Agent to S status.
Predominantly accusative languages are the mirror image of ergative ones also in
this respect and many of them have some mechanism that promotes the Patient (and
demotes the Agent).

4d. A O V ÷ (EA) SO V

(846) ren sha hu

man kill tiger
‘The man killed the tiger’
(847) hu sha (yu ren)
tiger kill (by man)
‘The tiger was killed (by the man)’ (Classical Chinese, Esa Itkonen, p.c.)

4d illustrates the simplest mechanism of Agent demotion or omission. There is no

marking on the verb and the primary means of passivization is the demotion of
Agent that appears in an oblique case. Furthermore, the Patient occupies the subject
slot in (847).
 325

4dd. A O V ÷ (EA) SO V-TR

(848) avan cannal-ai utai-t-an

he.NOM window-ACC break-W EAK-3SG.PAST
‘He has broken the window’
(849) cannal (paiyan ~l) utai-kk-ap pattatu
window.NOM (boy.INSTR) break-STRONG-PARTIC.PASS fall.3SG.PAST
‘The window has been broken (by the boy)’ (Tamil, Klaiman 1988: 45)

(850) háama-nm pée’wiye wewúkiye-ne

man-ERG he.shot.it elk-DO
‘The man shot an elk’
(851) wewúkiye ’ew-yíin hiwéeke
elk shoot-STAT 3NOM-was
‘The elk was shot’ (Nez Perce, Rude 1988: 547, 553)

(852) mangata nethakun yaramanin

child.NOM hit horse.ACC
‘The child hit the horse’
(853) yaraman neyikun (ma:nangan)
horse.NOM hit.PASS (child.ACC)
‘The horse was hit by the child’ (Lardil, Klokeid 1976: 552)

4dd illustrates what can be regarded as a typical passive (see, e.g. Siewierska 1984:
2, Shibatani 1985: 837 and Keenan 1985: 247). O is promoted in status, which is
accompanied by a demotion (or complete omission) of the Agent and the
passivization of the verb. Note that in (852)-(853) the Agent is demoted to the
accusative (cf. also Circassian, Karao and Warrungu below).

4ddd. A O V ÷ SO V-TR

(854) nainen näk-i poja-n

woman.NOM see-3SG.PAST boy-ACC
‘The woman saw the boy’
(855) poika näh-tiin
boy.NOM see-PASS.PAST
‘The boy was seen’ (Finnish)
326

4ddd illustrates a completely de-transitive passive. The verb is marked as passive,

which results in a complete syntactic omission of the Agent.

4e. A O V ÷ S V-TR

(856) annem kapi-yi aç-tv

mother.my door-ACC open-PAST
‘My mother opened the door’

(857) kapi aç-vl-di

door.NOM open-ANTIC-PAST
‘The door opened (Turkish, Haspelmath 1987: 2)

(858) devaki magu-vannu bacchiTTalu

Devaki.NOM child-ACC hid 3FEM
‘Devaki hid the child’
(859) magu bacchiTTukoNDanu
child.NOM hide.koLL(?).PAST.-3MASC
‘The child hid’ (Kannada, Amritavalli 1990: 288)

Type 4e possibly illustrates the most basic alternation of type 4. It is, however,
illustrated last, since it exemplifies the most complete degree of argument omission.
The events profiled in (857) and (859) are completely intransitive, i.e. they involve
one participant only. The two illustrated languages differ from each other in the
nature of the participant involved. In (857), the S argument refers to the patient,
while in (859) the participant referred to is more agentive. Since there is only one
participant involved, the appropriate notation is S V without further modifications
of S.

5. X Y V ÷ X Y V+x

Thus far, we have illustrated alternations that affect the number of arguments. In this
section, we proceed to cases in which the number of arguments is retained, but their
status is somehow affected. Also here, we proceed from simple to more complex
cases. Type 5 comprises cases in which the marking of arguments remains the same,
 327

and the only difference between basic and altered clauses lies in the verb
morphology. There are three different subtypes that are illustrated below. In
distinction from the general approach adopted in the present section, the division
below is merely semantically motivated.

5a. A O V ÷ A O V+TR

(860) mak’i-di »eni ibi

child-ERG water splash.PAST
‘The child splashed the water (perhaps involuntarily)’
(861) mak’i-di »eni ib-ali
child-ERG water splash-CAUS.PAST
‘The child splashed the water (purposefully and repeatedly)’ (Godoberi, Kibrik 1996: 128)

In 5a, changes in the verb morphology are associated with transitivity increase
(semantically). The causative marking stresses the higher degree of agency in (861).

5aa. A O V ÷ AE OA V+TR

(862) y-ttcu wqqzin aysum

3M S-eat dog.CST meat
‘The dog ate the meat’
(863) y-ss-ttc wryaz aqqzin
3M S-TRANS-eat man.CST dog
‘The man fed the dog’ (Berber, Guerssel 1986: 36)

5aa illustrates what was above labelled as transitivity rearranging alternation. There
are no changes in the marking or number of arguments in this particular case. What
makes it justified to label (863) as a transitivity alternation is that the number of
participants increases. Consequently, 5aa is here regarded as a transitivity increasing
(V+TR) alternation.

5b. A O V ÷ A O V-TR
328

(864) ngoah kohkoa oaring-kai

I grind coconut-these
‘I am grinding these coconuts’
(865) ngoah ko oaring
I grind coconut
‘I am coconut-grinding’ (Mokilese, Mithun 1984: 849, cited from Harrison 1976)

(866) waguja-õgu bana-Ø wawa-l

man-ERG water-ABS see-PRES
‘The man sees the water (on purpose, e.g. was looking for it)’

(867) waguja-õgu bana-Ø wawa-:ji-õ

man-ERG water-ABS see-ji-PRES
‘The man sees water (accidentally)’ (Yidiñ, Dixon 1994: 61)

As opposed to 5a and 5aa, the alternations presented above reduce the semantic
transitivity of clauses. In both Mokilese and Yidiñ, the marker employed is clearly
related to lower transitivity in general. Mokilese is a rather typical Oceanic
language, since the transitivity is typically marked on the verb. If the Patient is
eliminated or is indefinite, the transitive marking is omitted. The marker -ji- of
Yidiñ also expresses antipassive (cf. (661) and (662) and reflexive.

6. X Y V ÷ X Z V(-TR)

In 5, we illustrated some cases in which the argument marking is constant, but the
verb morphology is affected. Type 6 comprises alternations in which the marking
shifts from typical transitive to in various ways less transitive. We begin by
illustrating cases in which the Patient is affected and proceed from mild to more
severe alternations.

6a. A O V ÷ A AO V

(868) mú "tu §a " hk’úm

3.PAT 1.AGT kill
‘I killed him’
 329

(869) mu "l §a " hk’úm

3.AGT 1.AGT kill
‘I killed it (a fly)’ (Central Pomo, examples courtesy of Marianne Mithun)

6a illustrates a very interesting alternation. The resulting construction has two

explicitly marked A’s. Functionally, the alternation is similar to e.g. (375)-(376)
from Turkish, since also in Central Pomo prominent patients are referred to
differently from less important ones.

6b. A O V ÷ A SO V(-TR)

(870) ma-n c huy-bu kawwi

he-NOM(ERG) dog-ACC kicked
‘He kicked the dog’
(871) ma-n c huy kawwi
he-NOM(ERG) dog kicked
‘He kicks dogs (habitually)’ (Manipuri, Bhat & Ningomba 1997: 144)

(872) tílaaki i-nánan-a k’úsi-na

woman 3NOM-bring-PAST horse-OBJ
‘The woman brought the horse’
(873) tílaaki i-nánan-a k 'úsi
woman 3NOM-bring-PAST horse
‘The man brought a horse’ (Sahaptin, Rude 1997: 329)

In 6b, an explicitly marked Patient loses its marking, which makes it structurally
similar to S. Alternations of this kind are naturally restricted to languages in which
O is explicitly marked (i.e. predominantly accusative languages), since otherwise
the change can only be from O to E.

6c. A O V ÷ A EO V

(874) l’homme mange le pain

man eat.PRES.3SG ART bread
‘The man eats the bread’
330

(875) l’homme mange du pain

man eat.PRES.3SG PART bread
‘The man eats some bread’ (French)

(876) ngaa-§i waa § saay§-naa-ng

I-ERG bird.ABS hear-PRES-1SG
‘I hear a bird’ (non-agentive)
(877) ngaa-§i waa §-kaay § saay§-naa-ng
I-ERG bird-DAT hear-PRES-1SG
‘I listen to a bird’ (agentive) (Chepang, DeLancey 1981: 634)

In 6c, the demotion of O is more complete than in the two previous cases, since O
is demoted to E in status. French and Chepang illustrate nicely the difference
between accusative and ergative languages. In French, A is not marked explicitly,
whereas in Chepang A appears in the ergative. These kinds of alternation are
probably somewhat more frequently attested in accusative languages, since in many
ergative languages the Patient demotion is accompanied by an Agent promotion.

6cc. A O V ÷ A EO V-TR

(878) nyula nyaka-n wurripa

3SG.NOM see.P/P bee.ABS
‘He saw bees’
(879) ngaya nyaka-kali wurripa-wu katyarra
1SG.NOM see-ANTIP.P/P bee-DAT possum-DAT
‘I was looking for bees and possums’ (W arrungu, Tsunoda 1988: 606)

(880) nyulu wangarr maa-ni

3SG.NOM white person.ABS get-PAST
‘He married a white woman’
(881) nyulu wangaarr-gal maana-adhi
3SG.NOM white person-ADESS get-REFL.PAST
‘He got married with a white woman’ (Guugu Yimidhirr, Haviland 1979: 130)

6cc illustrates the other type of 6c. The difference lies in the explicit marking of the
verb in 6cc. It is perhaps worth mentioning that we have not come across a language
 331

in which the ergative marking of Agent would be retained in these kinds of case.
In the three previously exemplified types, the Patient has been demoted. The
most severe degree of demotion is illustrated by incorporation, as a result of which
the Patient loses its status as an independent argument, which typically de-
transitivizes the predicate. The Patient remains, however, a part of the clause. There
are two different types of incorporation (that can be regarded as relevant for our
purposes) illustrated below.

6d. A O V ÷ A VO

(882) (kamij c-n v) sipara ja-puhi-i

(1SG-ERG) axe 1SG-want-DYN
‘I want an axe’
(883) (kamij c-n v) ja-sipara-puhi-i
(1SG-ERG) 1SG-axe-want-dynamic
‘I want (it), the axe’ (Yanomam v, Aikhenvald & Dixon 1999: 350)

In Yanomamv, the incorporation of O does not result in other changes in the clause
structure. What is interesting here is that A retains its marking despite the decrease
in the number of independent arguments.

6dd. A O V ÷ SA V O

(884) tumg-e na-ntcwat-cn kupre-n

friends-ERG 3SG-set-TRANS net-ABS
‘The friends set the net’
(885) tumg-ct kupra-ntcwat-g’at
friends-ABS net-set-INTR
‘The friend set nets’ (Chukchi, Comrie 1973: 243f, cited from Skorik 1968: 267, glosses from
Payne 1997: 222)

What distinguishes 6dd from 6d is that the marking of Agent shifts from A to S.
This is due to the decreased number of nominal arguments.

6e. A O V ÷ OA O V
332

(886) chi-bashli-li-tok
2.ACC-cut-1.NOM-PAST
‘I cut you’
(887) chi-sa-banna-h
2.ACC-1.ACC-want-PRED
‘I want you’ (Choctaw, Davies 1986: 15, 65)

(888) ø-a-tsin kir eer-i oroz

3-IMPF-see jackal-NOM dog(ACC)
‘The jackal sees the dog’
(889) kir eer ø-a-tsin oroz
jackal(ACC) 3-IMPF-see dog(ACC)
‘The jackal sees the dog’ (Murle, Andersen 1988: 323)

6e illustrates the mirror image of 6a. In 6e the marking of Agent shifts from A to O,
because of which the clauses illustrated above have two explicitly marked O’s.

6f. A O V ÷ SA O V

(890) rasul(-lul) qata bullali-sa-r

Rasul(-ERG(1CL)) house.ABS(3CL) 3CL.build.DUR-PART-3SG
‘Rasul is building the house’ (Lak, Kazenin 1998: 101)

(891) krar-w kwr py y-wr-y

dog-ERG pig that.ABS OBJ-bite-PAST.SG.SUBJ
‘The dog bit the pig’
(892) krar kwr y-wr-y
dog pig OBJ-bite-PAST.SG.SUBJ
‘The dog bit pigs’ (Kanum, Donohue 1997: 8f)

In 6f, the Agent marking shifts from A to S. As a result, the Agent appears in a zero
case that in both exemplified languages (because of their basic ergative structure)
is identical to O. Consequently, it would, in principle, be possible to label 6f (and
6ff) as instances of O O V as well. However, what distinguishes these two types
from each other is the explicit accusative marking of O in (887) and (889): both
arguments are marked differently from S. On the other hand, in (890) and (892) it
 333

is more appropriate to regard the demoted A as SA, since the Agent appears in the
same case as S. In (887) and (889), the marking has to be considered O, since the
marking correlates with the semantic role referred to in transitive clauses.

6ff. A O V ÷ SA (O) V-TR

(893) àng-í pé tiq-chvng zvt-ò-‘

3SG-AGT basket one-CL weave-3.TR.N.PAST-N.PAST
‘He is weaving the basket’
(894) àng pé zvt-‘
3SG basket weave-N.PAST
‘He weaves baskets’ (Dulong/Rawang, LaPolla 2000: 285)

(895) cesar-nin-ra maria-nin wai-Ø rera-ke

Cesar-ERG-AS Maria-GEN field-ABS cut.tree-CMPL
‘Cesar cleared Maria’s field’
(896) ea-Ø-ra (enbix) rera-kaa-ke
1-ABS-AS (myself) cut-DETR-CMPL
‘I cut myself (by my own action)’ (Shipibo-Conibo, Valenzuela 1997: 28, 210)

In 6ff, the relevant change is the shift in the marking of Agent. As opposed to 6f, the
change is also signalled by the verb.

6g. A O V ÷ EA O V

(897) amma kuttiye atik’k’anam

mother.NOM child-ACC beat.anam
‘The mother must beat the child’
(898) ammak’k’c kuttiye atik’k’anam
mother-DAT child-ACC beat.anam
‘The mother wants to beat the child’ (Malayalam, Mohanan & Mohanan 1990: 45)

6g (as well as 6gg) illustrate cases in which the demotion of A is complete, since it
appears in a non-core E form. In 6g, there are no changes in the verb morphology.
334

6gg. A O V ÷ EA O V-TR

(899) lam cya bat kææva

child.NOM rice eat.PAST.ACT
‘The child ate rice’
(900) lam cya-šc vaha kævuna
child-DAT poison eat.PAST.PASS
‘The child (accidentally) ate something poisonous’ (Sinhala, W ijayawardhana et al 1995: 108)

6gg illustrates the ‘verbally marked variant’ of 6g. The changes in the argument
marking are similar in both cases.
In the cases presented thus far, the status of one argument only has been
affected. Below, we proceed to cases in which the marking of both arguments is
affected. The alert reader may object to the inclusion of the following three types in
the discussion here, since the type X Y V ÷ X Z V implies that only the marking
of one argument can be affected. However, in the cases below, the marking of either
argument shifts to the neutral S instead of E. The following represents a kind of
intermediate form between 6a-6g and the alternations illustrated in 7.

6h. A O V ÷ SA E V

(901) na va’ai e le fafine le pule= ~’oga i

PAST see ERG ART woman ART rule=school LD
l=o=na ofisa
ART=POSS=3SG office
‘The woman saw the principal at her office’
(902) na va’ai le fafine i le pule= ~’oga i le maketi
PAST see ART woman LD ART rule=school LD ART market
‘The woman saw the principal at the market (accidentally)’ (Samoan, Mosel & Hovdhaugen
1992: 424)

(903) ’ale-m ç’cg <c-r Ø-j-e-z <e

boy-ERG ground-ABS it-he-DYN-plough
‘The boy is ploughing the ground’
(904) ’ale-r ç’cg <c-m Ø-j-e-z <e
boy-ABS ground-ERG it-he-DYN-plough
 335

‘The boy is ploughing away at the ground’ (Circassian, Hewitt 1982: 158, cited from
Colarusso 1977: 132)

The Patient is here demoted from O to E. In both Samoan and Circassian, this
produces a change in the Agent marking. In both languages, A is promoted to S.
What is interesting is that the Patient is demoted from absolutive to ergative in
Circassian. However, since the demotion is to a case not typical of the Patient, we
have regarded the Circassian antipassive (or labile construction in the spirit of
Hewitt:ibidem) as an instance of SA E V-type. It is a mere coincidence that the
Patient in the antipassive appears in the same case as the Agent of typical transitive
clauses.

6hh. A O V ÷ SA E V-TR

(905) kapkap-en na to 'o 'i mangka

chop -ACT/TH/IMPF(ASP) ERG person ABS mango
‘The person will chop the mangoes’
(906) meN-kapkap 'i to 'o na mangka
ACT/AGT/IMPF-chop ABS person OBL mango
‘The person will chop some mangos’ (Karao, Brainard 1997: 87f)

(907) bama-õku gamu bidja-n

man-ERG water drink-NFUT
‘The man is drinking water’
(908) bama gamu-õku bidja-gali-n
man water-ERG drink-ANTIP-NFUT
‘The man is drinking water’ (W arrungu, Blake 1977: 25)

In 6hh, the Patient is also demoted from O to E, which results in a shift in the status
of Agent, and which is accompanied by a change in the verb morphology. Both in
Karao and Warrungu, the demotion is also to ergative (labelled as oblique by
Brainard).
336

6i. A O V ÷ EA SO V-TR

(909) ich hab-e den teller zerbrochen

I.NOM have.PRES-1SG ART.ACC plate break.PARTIC
‘I broke the plate’
(910) mir ist der teller zerbrochen
I.DAT be.PRES.3SG ART.NOM plate break.PARTIC
‘I broke the plate accidentally’ (German)

In 6i, the Agent is demoted to E, because of which the Patient appears in an S form.
In all the examples of this type we have come across, the verb is also marked.

6j. A O V ÷ OA AO V-TR

(911) mú "tu §a " hk’úm

3.PAT 1.AGT kill
‘I killed him’
(912) to " qadálmada mu "l
1.PAT hate 3.AGT
‘He hates me’
(913) mu "l qadála to " §ú "daw
3.AGT hate 1.PAT really
‘I really hate him’ (Central Pomo, examples courtesy of Marianne Mithun)

In the last type at issue here, the marking of both A and O has been affected. A is
marked as O, while O has inherited its marking from A. This is also the case in the
examples from Circassian, Karao and Warrungu. However, in (911)-(913) the
marking of both arguments is semantically determined. The patientive and agentive
arguments refer also in intransitive clauses to patientive and agentive participants,
respectively. Consequently, it is not possible to label neither of the arguments of
transitive clauses as S. We do not wish to label 6i as an instance of 7 either, though,
since the alternations illustrated below are such that neither of the two arguments
of transitive clauses appear in a form typical of basic transitive clauses.
 337

7. X Y V ÷ Z Z V(-TR)

Type 7 comprises cases in which the marking of both arguments is affected.

Differently from 6h and 6i, none of the arguments in the derived clause appears in
the same form as either argument of a basic clause. The resulting construction is the
same as is 6h-6j, but the differences in the basic clause structure make it necessary
to treat these as distinct types. Only languages with an obvious tripartite argument
marking pattern are possible candidates here. Consequently, the number of different
types is extremely low.

7a. A O V ÷ SA SO V-TR

(914) háama-nm pée’wiye wewúkiye-ne

man-ERG he.shot.it elk-DO
‘The man shot an elk’
(915) háama hi’wíye wewúkiye
man 3NOM.shot elk
‘The man shot an elk’ (Nez Perce, Rude 1988: 547, 552)

In 7a, both arguments of the derived construction are identical to S. The verb is not
explicitly marked as intransitive, but this is signalled by the verbal cross-reference.

7b. A O V ÷ S(A) E(O) V-TR

(916) piti-ya õa-tu ina

hit-PRES 1SG-ERG 2SG.ACC
‘I 'm hitting you’
(917) piti-li-ya õan yt ya in-ku
hit-li-PRES 1SG 2SG-DAT
‘I feel like to hit you’ (Pitta-Pitta, Blake 1979b: 207)

As opposed to 7a, the marking changes from A O to S E in 7b. The Agent appears
in the S form. The Patient is demoted from O to E.
338

8. X Z V ÷ X Y V(+TR)

In all the examples illustrated and discussed so far, the alternations have decreased
the transitivity of clauses. In the remainder of the section, we will illustrate some
transitivity increasing alternations that do not affect the number of arguments. The
original structure is less transitive than the resulting one. For convenience, we have
only taken account of cases in which the transitivization is complete, i.e. the
resulting structure is A O V (this means that the likes of E S V ÷ E O V are
ignored).

8a. A SO V ÷ AOV

(918) jin xk-a dar-Ø dist-ã

girl-OBL wood-DIR see.PAST-3PL
‘The girl saw some wood’
(919) jin xk-a dar-anã dist-Ø
girl-OBL wood-DAT/ACC see.PAST-Ø
‘The girl saw the wood’ (Balochi, Farrell 1995: 224)

(920) laka:h me Û •]]h-ê §

hen jacamim peck-PAST
‘A hen pecked a jacamin bird’
(921) laka:h •]]h-ê § me Û-v:y§
hen peck-PAST jacamin-O.TOPICAL
‘As for the jacamin bird, a hen pecked it’ (Dâw, Martins & Martins 1999: 263f)

In 8a, the marking of Patient changes from S to O. According to Farrell (ibidem),

the dative marking appears only if the emphasis is on the Patient. The default choice
is that the Patient is unmarked. The situation is very similar in Dâw. The alternation
type illustrated here is very problematic, since usually constructions like (919) and
(921) are rather regarded as typically transitive, while (918) and (920) are less
transitive versions of them. Consequently, these cases rather illustrate a transitivity
decrease. These kinds of alternation tend to be conditioned by definiteness of
arguments. If we come across a language in which A S V denotes semantically less
transitive events that can be transitivized by changing the marking of object, we can
 339

justly argue for the transitivizing nature of these alternations. What makes it
marginally possible to regard (919) and (921) as transitivizing alternations is the
lower frequency of occurrence of (919) and (921) (cf. above). Despite this, the status
of these kinds of alternation remains very vague at best.

8b. A EO V ÷ AOV

(922) mies rakas-ti koira-a-nsa

man.NOM love-3SG.PAST dog-PART-3POSS
‘The man loved his dog’
(923) mies rakas-ti koira-n-sa kuoliaaksi
man.NOM love-3SG.PAST dog-ACC-3POSS to.death
‘The man loved his dog to death’ (Finnish)

(924) sa-niryúúy yíva

3SG-desire 2SG.DAT
‘S/he has desire towards you’
(925) sa-niryúu-jíy
3SG-desire-2SG
‘S/he desires you’ (Yagua, Payne 1985: 36)

What distinguishes 8b from 8a is that in 8b (and 8bb) the marking of Patient is E

and not S. This kind of alternation is clearly transitivizing, since less than perfectly
transitive clauses are completely transitivized. The status of this alternation as a
genuine transitivizing alternation is considerably higher than that of (8a), since (923)
and (925) illustrate verbs in untypical environments. The derivation cannot be from
(923) to (922) or from (925) to (924).

8bb. A EO V ÷ A O V+TR

(926) nii ong-jo pehkoro iirong-ohna-na

1SG DEM.M-PURP boy get.angry-1S a.PRES.PROG-F
‘I am angry for the sake of this boy’
(927) nii ong pehkoro iirong-ee-uhna-na
1SG DEM.M boy get.angry-APPL-3O.1A.PRES.PROG-F
‘I am angry with this boy’ (Motuna, Onishi 2000: 132)
340

(928) kaie mutau eni tirausis onene

3SG 3SG.real.defecate OBL trousers POSS.man.3SG
‘He defecated on his shorts’
(929) kaie mutau-ti tirausis onene
3SG 3SG.real.defecate-TRANS shorts POSS.man.3SG
‘He shat his shorts’ (Paamese, Crowley 1982: 191)

Also in 8bb, semantically rather intransitive events are transitivized. Examples from
Motuna are very similar to (922) and (923). The main difference is that in Motuna,
the verb is applicativized. Examples (928) and (929) illustrate a case in which a non-
core argument (and a participant) is promoted to the core.

8c. SA O V ÷ AOV

(930) wa mási a-ka-i-e

man boy 3SGU-see-3SGA-DECL
‘The man saw the boy’
(931) mási wá-má a-ke-i-e
boy man-ERG 3SGU-see-3SGA-DECL
‘The man sees the boy’ (Fore, Foley 1986: 172)

(932) ea-Ø-ra yapa-Ø pi-kas-ai

1-ABS-AS fish-ABS eat-DES-INC
‘I want to eat fish (referential or non-referential)’
(933) e-n-ra yapa-Ø pi-kas-ai
1-ERG-AS fish-ABS eat-DES-INC
‘I want to eat the fish (referential only)’ (Shipibo-Conibo, Valenzuela 1997: 197 91)

8c is in a way the mirror image of 8a, since the clause is transitivized by attaching
an ergative affix to the Agent. Since we are dealing with two predominantly ergative
languages, the marking of O remains constant.

8cc. SA O V ÷ A O V+TR

91
The direction in desiderative constructions of Shipibo-Conibo is from (932) to (933). The former
is the expected pattern, whereas (933) is possible, if the patient is highly referential.
 341

(934) nii tuu haa-mu-u-ng

1SG water want-1S o-NR.PAST-M
‘I want water’
(935) ong tuu ni-ngi haa-jee-unho-ng
DEM.M water 1SG-ERG want-APPL-3O.1A.PRES:PROG-M
‘I need this water (for some purpose)’ (Motuna, Onishi 2000: 134)

8cc differs from 8c in that the verb is explicitly marked as transitive.

8d. E O V ÷ A O V+TR

(936) marasin-an mamam p-waca-k-m-tv-t

medicine V SG-OBL sore.VII.SG VII SG.O-become.small-IRR-VII.SG-become-PERF
‘The sore healed because of the medicine’
(937) marasin mamam p-n-tar-waca-k-m-tv-t
medicine V SG sore VII SG V II SG O - 3SG A-CAUS-become small-IRR-V II SG -
become-PERF
‘The medicine healed the sore’ (Yimas, Foley 1991: 300)

In 8d, an oblique Agent is promoted to A status, in addition to which the verb is

transitivized.

9. Z Z V ÷ X Y V(+TR)

Alternations exemplified below illustrate cases in which morphologically

intransitive clauses are completely transitivized. The label Z Z V refers to cases with
two arguments neither of which is a typical Agent or a Patient. Formally, either
argument in the transitivized clause may correspond to an argument in the less
transitive clause, but the functional differences are very obvious and explicitly
distinguish between seemingly identical cases.

9a. SA E V ÷ AOV

(938) fakalilifu a ia ke tau momotua

respect ABS he to PL old.PL
342

‘He respects the old people’

(939) fakalilifu e ia e tau momotua
respect ERG he ABS PL old.PL
‘He respects the old people’ (Niuean, Seiter 1980: 336)

In (938), the Agent appears in the absolutive and the Patient in an oblique case. The
Patient in (939) and the Agent in (938) are structurally identical. The difference is
that in (938) the use of absolutive is structurally motivated, whereas in (939) the
form is semantically conditioned, since the Patient of typical ergative systems is
zero marked. The Agent appears in the ergative in (939).

9aa. SA E V ÷ A O V+TR

(940) e alofa le teine i le tama

GENR love ART girl LD ART boy
‘The girl loves the boy’
(941) e alofa=gia ‘it~tou e le nu’u
GENR love=ES 1INCPL ERG ART village
‘W e are well treated by the village’ (Samoan, Mosel & Hovdhaugen 1992: 731)

Type 9aa illustrates the other subtype of 9a in which the verb is transitivized.

9b. S (E) V ÷ AOV

(942) kua mohe a ia he fale

PERF sleep ABS he in house
‘He has slept in the house’
(943) kua mohe e ia e timeni
PERF sleep ERG he ABS floor
‘He has slept on the floor’ (there is some effect on the patient) (Niuean, Seiter 1980: 63)

The main difference between 9a and 9b (as well 9bb) is that in 9b the obliquely
marked argument is also ‘functionally oblique’. It is not an integral part of the event
denoted. However, as illustrated above, it might be promoted to core status, in
which case the whole clause is transitivized.
 343

9bb. S (E) V ÷ A O V+TR

(944) make matin-do (nana ipa-yave)

boy wash-IND (OBL river-DEF)
‘(The) boy is washing (in the river)’
(945) make matin-na ipa-yave
boy wash-APPL river-DEF
‘(The) boy is washing in the river’ (W arembori, Donohue 1999b: 9)

9bb illustrates a rather typical applicative derived from an intransitive clause. The
transitivization is also signalled on the verb, which makes the omission of the
Patient ungrammatical.
The typology illustrated above is based primarily on structural differences
between alternations. As noted (and illustrated) above, the basic structure of
languages makes a very significant contribution to the nature of alternations. A
concrete example of this is provided by the differences between 4ccc and 4ddd.
Since the Agent is the explicitly marked argument in ergative languages, the
marking of Agent can shift from A to S only in these systems. In accusative
languages, the function expressed may be the same, but differences in agency cannot
be marked by manipulating the marking of Agent in the same way. Similarly, in
accusative languages, the Patient marking can shift from O to S and E, while in
ergative languages only the latter is possible.
A further consequence of the structural emphasis in our typology is that
languages differ crucially from each other in what kinds of semantic or functional
alternation are associated with the structural ones (cf. also Lazard 1998: 164 among
others). For example, the alternation A O V ÷ A E V is in some languages
conditioned by (in)definiteness, whereas in others it expresses a shift from transitive
to less transitive events in general. Conversely, a certain functional alternation can
be expressed by different alternations depending on the structure of a language. We
can naturally draw some generalizations, but we are not likely to find an alternation
type that exclusively related to the expression of a single function cross-
linguistically. Alternations that increase the number of arguments of originally
transitive clauses are the most likely candidates. This is due to the fact that the only
possible alternations of this kind are causatives and applicatives. Furthermore, as
344

illustrated in 3, causatives and applicatives differ crucially from each other in the
nature of the added participant. This is also directly reflected in the consequences
the alternation has for the participants in an event (this was discussed more
thoroughly above). However, the structural differences between these alternations
are not so obvious, as the functional differences may imply and, as has been shown,
for example, in 3aa and 3aaa. The result of the alternations may be the same. Hence,
we do not wish to argue for a perfect correlation of a semantics and structure here,
either. What is, though, perhaps noteworthy is that the most typical alternation
associated with the expression of causatives (of transitives) is rather marginally
employed in the expression of applicatives and vice versa. Put concretely, this
means that 3g is very likely to express causatives and 3aa applicatives, but there are
also numerous exceptions. On the other hand, type 4e expresses a great variety of
semantically and functionally different alternations. This variety of functions
follows from the nature of the eliminated argument that can refer to both agent and
patient in a given event, in addition to which the rationale behind the omission may
vary greatly.

5.3. Semantic-functional typology of transitivity alternations

In the previous section, we illustrated a rather detailed structural typology of

transitivity alternations. The semantic or functional nature of the alternations was
largely ignored above, even if some generalizations were briefly discussed. The
topic of the present section is the exact opposite of this. We will here discuss the
functional rationale behind the alternations. Put concretely, this means that we aim
at showing what kinds of semantic (ontological or conceptual), pragmatic (defined
in a rather loose sense covering most alternations that do not affect the ontology of
events) or purely (or at least primarily) structural features can result in a transitivity
alternation. The relation between morphosyntax and semantics is assumed to be
iconic here, which means that an alternation that decreases the semantic transitivity
of events intransitivizes the structure. The structural nature of the alternations is
rather (yet not completely) irrelevant. It is naturally of the utmost importance that
the derived constructions differ somehow from the original ones. Structurally
identical alternations (e.g. two functionally different instances of A O V ÷ E O V)
 345

are subclassified only on the basis of the related functions. In some cases,
semantically different alternations are distinguished also morphosyntactically,
though. This is important, if we try to justify differences between two functionally
very similar alternations. Furthermore, alternations related to the same function are
not distinguished irrespective of how significant the structural differences are.
Usually, we exemplify the alternations in light of a couple of examples simply to
illustrate their morphosyntactic variety.
In the following, alternations are divided into three main types based on the
conditioning motivations and on how the alternations are expressed (are they
expressed directly or indirectly). The alternations are here labelled as direct, indirect
and structurally conditioned alternations. The types presented are not hierarchically
organized nor do they illustrate different types of one basic type. Rather, they
illustrate different facets of (linguistic) transitivity. The first type comprises directly
observable semantically (in a broad sense) or pragmatically determined alternations.
The saliency discriminates between direct and indirect alternations. In the latter
case, changes in transitivity are reflected only indirectly at the level of
morphosyntax. This means that the morphosyntax of clauses as such remains
unaffected, and the semantic changes are reflected indirectly. A typical example of
this is presented by restrictions on passivization. The conditioning principles are the
same in the two first types. Structurally motivated alternations differ from others in
that they are not motivated by semantic or pragmatic changes, but they are a
grammatical necessity. They are expressed by the same mechanisms as direct
alternations. By including indirect and structurally motivated alternations to our
typology, we hope to be able to study the notion of transitivity in more detail.
Indirect alternations show that directly observable changes in the clause structure
are not the only manifestation of transitivity, but differences in transitivity are
expressed also in other ways (cf. also indirect marking in 4.3.). Structurally
conditioned alternations, on the other hand, emphasize the fact that changes in
structural transitivity do not necessarily reflect any ontological, conceptual or
pragmatic changes. Because of their highly structural nature, the latter kinds of
alternation are the least universal and most language-specific. Together, the
illustrated types should provide a rather thorough overview of transitivity.
346

5.3.1. Direct transitivity alternations

Direct transitivity alternations exemplify what is usually labelled as transitivity

alternations. In these cases, semantic or pragmatic changes related to transitivity are
directly reflected in the clause structure employed in the description of events.
Based on the underlying motivation (as well as their consequences), direct
alternations can be divided into minus-transitivizing and (plus-)transitivizing
alternations. The former is further subdivided into intransitivizing and de-
transitivizing alternations. This is based on whether an alternation is motivated by
semantics or pragmatics (pragmatics refers here to alternations that do not affect the
semantics of events). Transitivizing alternations are also divided into two based on
similar criteria (see below for details). In the following, we illustrate all of these
types in detail in light of cross-linguistic data.

5.3.1.1. Minus-transitivizing alternations

In a nutshell, as minus-transitivizing are here labelled directly observable

morphosyntactic alternations motivated by reduced transitivity. Typical features that
result in a minus-transitivizing alternation include decreased agency and
affectedness, in addition to which (in)definiteness of arguments is also of the utmost
importance. The consequences of minus-transitivizing alternations vary. In the most
radical cases, the shift is from typical transitive to completely intransitive clauses.
Moreover, there are also numerous intermediate cases. Defining an alternation as
minus-transitivizing is here primarily semantically conditioned, which means that
in principle any structural alternation determined by reduced semantic transitivity
is taken account of. Alternations that affect the number or case marking of
arguments are the most obvious examples of these. Less obvious, and hence
somewhat more problematic, are cases in which changes are signalled merely by the
verb (see discussion in 5.1).
In this study, two major types of minus-transitivizing alternations are
distinguished based on whether the alternations are conditioned by semantic
transitivity or not (semantic transitivity covers ontological transitivity and some
aspects of conceptual transitivity). The former type is referred to as intransitivizing
 347

alternations, whereas the latter kinds of alternation are labelled as de-transitivizing.

The result of an intransitivizing alternation is clearly a less transitive clause,
whereas de-transitivized clauses can profile highly transitive events. Both types are
further subclassified depending on whether an alternation affects the number of
participants or arguments or whether it only has some minor consequences for the
clause structure. We will illustrate the four types of minus-transitivizing alternations
in light of concrete examples from a variety of languages below. We also discuss the
specific underlying motivations of individual types in detail.

5.3.1.1.1. Intransitivizing alternations

We begin by illustrating intransitivizing alternations. Intransitivization is here

understood as a semantically conditioned structural intransitivization. This means
that only cases in which we can explain the structural decreases in transitivity by
referring to changes in semantic transitivity, are included in the typology.
Alternations that merely intransitivize the structure are excluded from in the present
section (they are discussed below). Of the two possible types of intransitivizing
alternations, those that affect the number of participants have the most drastic
consequences for the transitivity of events, which is also reflected structurally. In the
other possible case, minor transitivity features are affected. We begin by illustrating
and discussing alternations that affect the number of participants, since these are the
most obvious examples of minus-transitivizing alternations.

5.3.1.1.1.1. Alternations affecting the number of participants

Since typical transitive events involve two core participants (an agent and a patient),
there are in principle only two possible kinds of intransitivizing alternation affecting
the number of participants. Either the agent or the patient is eliminated. For
example, ‘he painted a house’ is a transitive event, whereas ‘he fell’ and ‘he ran’ are
intransitive events. Differently from typical transitive events, ‘he fell’ does not
involve an agent, while ‘he ran’ lacks a patient. The actual number of these kinds
of alternation is, however, higher than two, since the rationale behind the
alternation also makes a contribution. Alternations conditioned by inherent
348

(in)transitivity of events will not be illustrated or discussed in detail in the present

section. The emphasis lies here on cases in which the basic and the manipulated
event clearly have common features. Inherently transitive events are excluded from
the description, since they necessarily involve two participants neither of which can
be eliminated. We begin by illustrating agent-omitting alternations.
There are basically two agent-omitting alternations. The first type is illustrated
by anticausatives, whereas resultatives present the other possible type. The agent has
been omitted from the cognitive structure of events in both cases, but the rationale
behind the omission varies. In anticausatives, the agent is eliminated at the
ontological level, wherefore it is completely absent during the whole duration of an
event (see also Siewierska 1984: 77 and Haspelmath 1987: 7). For example, a
spontaneous breaking of a window is not instigated by an agent, because of which
the agent cannot be present at any stage of the event. In rresultatives, on the other
hand, the agent is omitted from the structure of the resultative event (or rather state)
only. The agent has already carried out a transitive action and it is no longer an
integral part of the event. This reflects the inherent non-agency of states. As opposed
to anticausatives, resultatives are thus originally transitive events, the intransitivity
of which follows from focussing exclusively on the resulting state of an event.
Typical examples of anticausatives and resultatives (respectively) are given below,
cf.

(946) devuška sloma-la palk-u

girl.FEM break-PAST.FEM .SG stick-ACC
‘The girl broke the stick’
(947) palk-a sloma-la-s’
stick.FEM-NOM break-PAST.FEM.SG-ANTIC
‘The stick broke’ (Russian, Haspelmath 1987: 2)

(948) nu õan tad ã kalan-me loko-d’oro-n

he.NOM there pot-ACC hang-PRES-3SG
‘He is hanging a pot there’
(949) tad ã kalan lok ã- a-d’ara-n
there pot.NOM hang-STAT-PRES-3SG
‘A pot is hanging (hangs) there’ (Evenki, Nedjalkov & Nedjalkov 1988: 242)
 349

In (946), the breaking of the stick is due to a directed action by the agent.
Consequently, the profiled event involves two participants and is transitive. In (947),
the agent is completely absent during the whole event. The stick breaks
spontaneously by itself without any involvement of an agent. The only difference
between (946) and (947) is the absence of agent. Both events result in a directly
observable change-of-state in the patient. Examples from Evenki are different in this
respect. In (949), the hanging of a pot on the wall results from a transitive event
denoted in (948). Example (949) only profiles the state resulting from (948). The
state is clearly intransitive, since the agent no longer is an integral part of the
denoted event. It is usually not possible to refer to the agent explicitly in
resultatives, even if there are exceptions. Nedjalkov & Jaxontov (1988: 51ff) state
that ‘agentive objects’ can be used in resultatives in many Indo-European languages
including German, Russian, Hindi, Norwegian and Armenian, whereas this is quite
untypical of languages outside of the Indo-European language family. The latter
kinds of language are perhaps more natural, since the agent is not an integral part
of the meaning of resultatives. As regards the meaning of anticausative and
resultative in general, we may conclude that (946) and (947) describe two
completely different events (one of which involves an agent and the other does not),
while (948) and (949) are more closely related to each other. (948) describes the
event as a whole, while (949) stresses one facet of it. Notice that the complete
absence of an agent in anticausatives implies that the given event has not been
instigated by a salient agent whose action we can directly observe. Also in
anticausatives, the change-of-state in the patient is due an implicit cause, such as
wind (e.g. in the case of opening a door), fire (burning), diseases (dying) etc. For
example, melting follows from high temperatures. This is, however, an inherent
property of melting and we do not have to express it explicitly, but it is more natural
to construe this event as anticausative. It is also possible to express an implicit cause
explicitly (examples follow below).
We can distinguish between two different resultative constructions on the
basis of the (in)transitivity of the preceding event. In the case of anticausatives this
is not possible, since anticausatives are completely intransitive. Consequently, we
do find languages with two different resultative constructions sensitive to
transitivity, cf.
350

(950) saa biu-ti a+i sele i+na dela ni teepeli

ASP be.put.PASS DET knife LOC+DET top of table
‘The knife has been placed on the table’
(951) saa biu a+i- sele i+na dela ni teepeli
ASP be.put DET knife LOC+DET top of table
‘The knife is in place on the table’ (Boumaa Fijian, Dixon 1988: 207) 92

(952) die tür ist geöffnet

ART.NOM door be.PRES.3SG open.PARTIC
‘The door is open’ (transitive)
(953) die tür ist offen
ART.NOM door be.PRES.3SG open
‘The door is open’ (intransitive) (German)

Examples from Boumaa Fijian illustrate a rather typical scenario. (950) is a

‘transitive resultative’, which is reflected in the passive marking of the verb.
Passives derived from transitive clauses imply that the change-of-state in the patient
follows from a targeted action by an agent. Hence, it is rather natural that clauses
like (950) are interpreted as profiling states resulting from transitive events. On the
other hand, (951) is more appropriate, if the speaker is not aware of any transitive
event that would have resulted in the state profiled. Consequently, the construction
implies intransitivity. German represents the other possibility of distinguishing the
two resultatives noted above. (952) describes a ‘transitive state’. The door has been
opened by an agent whose identity is irrelevant as regards the state at issue. Only the
state of an originally transitive event is focussed on. Example (953), on the other
hand, denotes a mere state without any reference to a preceding transitive event. It
is appropriate, if we do not know whether the door has been opened by someone or
whether it has opened spontaneously. In German, the difference between ‘transitive’
and ‘intransitive states’ is expressed by substituting an intransitive verbal
construction with a copula construction. In (953), the infinite verb of (952) is
replaced by an adjective not directly derived from the transitive verb. As a result, the
actionality is completely backgrounded. Notice that only (950) and (952) illustrate
genuine transitivity alternations. The original event involves two participants,

92
I thank Ritsuko Kikusawa for the glosses.
 351

whereas the number of arguments is reduced in resultatives due to the agent

omission. Intransitive resultatives, on the other hand, rather express a change in the
perspective that shifts from process to state. The number of participants remains the
same, and we simply change the viewpoint. Since the number of participants is one
in both cases, intransitive resultatives are not regarded as transitivity alternations in
the present study (cf. (348)-(351) above).
The obvious differences in motivation of anticausatives and resultatives are
morphosyntactically neutralized in many languages, cf.

(954) y-rzem wryaz tawwurt

3MS-open man.CST door
‘The man opened the door’
(955) t-rzem tewwurt
3MS-open door.CST
‘The door opened/the door is open’ (Berber, Guerssel 1986: 48)

(956) orang itu buka pintu itu

person that open door that
‘That person/man opened the door’
(957) pintu itu (tiba-tiba) ter-buka
door that (suddenly) ter-open
‘The door suddenly opened/is open’ (Malay, examples courtesy of Foong Ha Yap)

Berber and Malay differ from each other in that in Berber the verb rzem is labile and
the number of arguments expressed is the main indicator of transitivity. In case only
one argument is expressed, anticausative and resultative are both felicitous readings.
In Malay, on the other hand, anticausative and resultative are both marked by the
verb prefix ter-. The adverb tiba-tiba ‘suddenly’ stresses the processual nature of the
event and the anticausative is the more plausible reading. Without this adverb, both
anticausative and resultative are equally possible.
It was shown above that the elimination of agent may be motivated in one of
two ways. Below, we will illustrate reflexives that illustrate the only transitivity
alternation that genuinely omits the patient. Before proceeding to details, it has to
be stressed that the general status of patient omitting alternations is lower than that
352

of agent omitting ones. In reflexives, the semantic role of the patient remains an
integral part of the event, even if the independent referent is omitted. Alternations
that completely eliminate the patient do not seem to exist. This may result from the
fact that patients make a very important contribution to the overall nature of
transitive events. For example, an event is an instance of ‘killing’ only if there is a
patient that dies. The nature of the action by the agent is less relevant.
Reflexives are semantically transitive events (i.e. they involve two semantic
roles) that involve one participant only. The only difference between transitive and
reflexive events is that in reflexives the semantic roles of agent and patient are
properties of a single participant. The affectedness follows from a transitive event,
which becomes obvious, if we compare ‘he killed himself’ to ‘he dies’. The
affectedness follows from a targeted action only in the former case. Since the sole
participant of a reflexive bears two roles, the eliminated participant could be either
an agent or a patient. The semantic nature of the reflexives points, however, to the
analysis proposed here. Similarly to transitive events, the agent is intending a
reflexive event to occur. This speaks for the elimination of the patient. A further
important feature is the passive nature of a typical patient. Typical patients are
completely passive participants that register the effect of an event. The sole
participant in reflexives is far from being a mere passive participant, since it is
volitionally causing itself to undergo a change-of-state. The occurrence of reflexive
events completely relies on the action by the sole participant. This is also reflected
in the fact that inherently passive participants (i.e. inanimate entities) cannot be the
sole participants of reflexives. For example, the event ‘the flowers died’ cannot be
interpreted reflexively, whereas ‘s/he died’ (understood simply as an event in which
a human being dies irrespective of transitivity) allows both anticausative and
reflexive readings. The emphasis on the agency distinguishes reflexives from closely
related anticausatives. As a result, it is more natural to conclude that the sole
participant of reflexives is primarily an agent that differs from a typical agent in that
it is also directly affected by the event in question.
The ‘dual’ nature of reflexives manifests itself also morphosyntactically.
There are three main types of reflexive construction that stress different aspects of
reflexive semantics, cf.
 353

(958) ti hi xib-áo-b-á-há
1 3 hit-TELIC-PERF-REMOTE-COM PLETE CERT
‘I hit him’
(959) ti ti xib-áo-b-á-há
1 1 hit-TELIC-PERF-REMOTE-COM PLETE CERT
‘I hit myself’ (Pirahã, Everett 1986: 216)

(960) i te amji p vtā÷

1SG ERG.PAST REFL defend
‘I defended myself’ (Timbíra, Rodrigues 1999b: 195)

(961) a-b c a-r Ø-g’a-y-c-8aa pw-a p-s

3-OBL 3-ABS 3-HOR-3-NONPRES-see-PAST-AFF
‘He saw him’ (HOR=horizon of interest)
(962) a-b c yar c-r z-y-c-8aa pwc-ž-a p-s
3-OBL self-ABS REFL-3-NONPRES-see-self-PAST-AFF
‘He saw himself’ (Kabardian, Colarusso 1992: 137)

(963) ama-:lu untawu yuku-õku aru-

man-ERG woman.ABS stick-INSTR hit-PAST
‘The man hit the woman with the stick’
(964) ama (uluma-uluma) yuku-õku ari-:ni-n
man.ABS (3SG-REFL) stick-INSTR hit-REFL-PAST
‘The man hit himself with a stick’ (Uradhi, Crowley 1983: 340)

(965) era chuchillu-ra ber-ki de-wei-ate

I knife-INSTR one-ACC 3SG-cut-1SGPAST
‘I cut someone else with a knife’
(966) era chchillu-ra wei-ate
I knife-INSTR cut-1SGPAST
‘I cut myself with a knife’ (Leko, van de Kerke 1998: 197)

Examples from Pirahá, Timbíra, Kabardian and English (see the translations)
illustrate reflexives derived by replacing the Patient of transitive clauses by a
(reflexive) pronoun or marker. The languages differ from each other in whether the
reflexive marker is similar to typical Patient pronouns or whether there the language
uses a specific reflexive marker in order to express reflexivity. In Pirahá, the
354

reflexivity is expressed by ‘doubling’ the Agent pronoun. In English, the element

in question is a true reflexive pronoun, whereas in Timbíra and Kabardian the
coreference of agent and patient is marked by a general reflexive particle. Examples
(958)-(962) illustrate transitive reflexives. In Kabardian, the reflexivity is also
signalled by the verb. This is characteristic of the second type in which reflexivity
is marked by two mechanisms. Uradhi is very similar to Kabardian in this respect,
but the reflexive pronoun is optional in (964). Furthermore, reflexive is structurally
less transitive in Uradhi, since the ergative marking of Agent is omitted.93 Uradhi
illustrates a rather typical intransitive reflexive construction (i.e. the third type). In
Leko, reflexivity is expressed by eliminating the Patient, which intransitivizes the
verb. Reflexivity is, however, not marked by a reflexive affix of any kind. This kind
of reflexivity is found also in numerous other languages, including English (he
shaves, he bathes). In the latter kind of case, there is no explicit marking of
reflexivity and the reflexive cannot be viewed as a distinct construction of its own.
Reflexives illustrated above emphasize different facets. Transitive reflexives
focus on the number of semantic roles present. The substitution of the Patient by a
co-referential (reflexive) pronoun suffices to signal the coreferentiality of agent and
patient. On the other hand, the likes of (964) and (966) stress the number of
individual participants. The semantics of events suffices to distinguish reflexives
from intransitive events. Languages like Kabardian (and also Uradhi) emphasize
both relevant aspects explicitly. These constructions are rather uneconomical and
are very likely in the minority cross-linguistically. They are simply redundant, since
simpler mechanisms suffice to explicitly distinguish reflexives from transitive
constructions. These obvious typological differences result from the dual nature of
reflexives. Reflexivization does not cause any changes in the ontological transitivity
of events, since reflexives involve two semantic roles. This kind of typological
variety is not attested in anticausatives, since their ontological transitivity allow
them to be expressed by one basic construction type only (i.e. A O V ÷ SO/O V-
TR).94

93
This is very likely due to the optionality of the reflexive pronoun.

94
W e have not regarded lexical and morphosyntactic anticausatives as distinct types.
 355

Reflexives stress the importance of the involvement of two distinct

participants for high transitivity. This is so, even if there are also transitive
reflexives, as shown above. A couple of further examples related to reflexivity are
illustrated below, cf.

(967) kaig-ni-a-a
hungry-claim-IND.TRANS-3SG/3SG
‘She says he is hungry’
(968) kaig-yuke-u-q
be.hungry-think-IND.INTR-3SG
‘She thinks she’s hungry’ (Yup’ik, Mithun 2000: 107)

(969) cnan ctlon lelu-nimen-nin

he.ERG he.ABS shaved his moustache
‘He i shaved his j moustache’
(970) ctlon lelu-nimet-g §i
he.ABS shaved (AOR) the moustache
‘He shaved the moustache (his own)’ (Chukchi, Kozinsky et al 1988: 687)

(971) ngatha wirnta-rna-pula jina

1SG.NOM cut-PAST-REFL foot
‘I cut myself in the foot’ (Panyjima, Dench 1991: 160)

The event profiled in (967) involves two distinct participants and the construction
employed is consequently transitive. (968) illustrates an explicit reflexive in which
the cross-reference of the verb is intransitive. Events denoted in (969) and (970)
both involve two participants, but only in (969) the distinct non-linguistic entities
are referred to by independent arguments. Consequently, (969) is transitive and
(970) intransitive. In (971), the verb is marked reflexively in order to emphasize that
the patient is a part of the agent.
Examples (958)-(971) illustrate reflexives that can be labelled as typical. The
agent is co-referential with the patient and there is only one distinct participant, yet
two semantic roles. Moreover, there are reflexives whose use is not determined by
the number of distinct participants, but by the importance of the notion of self. An
example is given below, cf.
356

(972) à:ng-í shvmø shvt-nò-‘

3SG-AGT mosquito kill-3.TR.N.PAST-N.PAST
‘He is killing a mosquito’
(973) àng shvmø shvt-shì-‘
3SG mosquito kill-REFL/MID-N.PAST
‘He is killing a mosquito (on him)’
(974) àng nø àng vdór-shì-‘
3SG TOP 3SG hit-REFL/MID-N.PAST
‘He is hitting himself’
(975) àng nø àng vd Çr-shì-‘
3SG TOP 3SG hit-REFL/MID-N.PAST
‘He is hitting his own (child, etc.)’ (Dulong/Rawang, LaPolla 2000: 292f)

All the events profiled above involve two distinct participants. As opposed to typical
reflexives, the number of participants is not affected. Despite this, the verb
morphology is reflexive in (972) and (973). According to LaPolla (ibidem),
reflexivity is motivated by overlapping of roles. Differently from typical reflexives,
in which one argument refers to both agent and patient, the roles are agent and
benefactive in (973). (974) and (975) (respectively) exemplify what LaPolla labels
as direct and indirect reflexives. In the latter case, the patient must not be co-
referential with the agent, but it has to closely related to the actor. As can be seen,
the differences between direct and indirect reflexives are expressed by tonal
differences. The reflexive marker itself is the same in both cases.

5.3.1.1.1.2. Alternations affecting certain aspects of semantic transitivity

In this section, we will illustrate alternations that only affect individual transitivity
features. The most important features of this kind are agency and affectedness
parameters. Also the concreteness of events makes a very significant contribution.
The number of affected features naturally varies, which correlates roughly with the
degree of changes in transitivity. We begin by examining cases in which the overall
transitivity of events as such is affected and end the presentation with cases in which
a single feature conditions the alternations.
The most obvious deviations from the basic transitive event are illustrated by
changes in the inherent transitivity of events. In many languages, only events the
 357

inherent transitivity of which is very high can appear in transitive case frames.
Examples of this are probably found in some form in all languages. A couple of
examples are given below, cf.

(976) mama-Ø Ø-aseneb-s saxl-s

father-NOM 3SG.OBJ-build-3SG.SUBJ house-DAT
‘Father builds the house’
(977) mama-s mo-s-c'on-s saxl-i
father-DAT PRV-3SG.SUBJ-like-3SG.OBJ house-NOM
‘Father likes the house’ (Georgian, Testelec 1998: 30)

(978) w 4 b| chá-b ‘i nòng-pò LE

I BA tea-cup make-broken PFV/CRS
‘I broke the teacup’
(979) *t~ b| xi|o m ~o ài
3SG BA small cat love
‘S/he loves the kitten’ (Mandarin Chinese, Li & Thompson 1989: 466f)

(980) zakurr-ek zikindu du etxe-a

dog-ERG dirty/PARTIC AUX/3SGI/3SGII house-ABS/DEF
‘The dog dirtied the house’
(981) piro-a gan zaiote auzu-ei
duck-ABS/DEF go.PARTIC AUX/3SGI/3PLIII neighbour-PL/DAT
‘The neighbours lost their duck’ (Basque, Lazard 1998:45, cited from N’Diaye 1970: 52f)

All the examples above illustrate the less transitive marking of experiencer
constructions. The profiled events are inherently less agentive and they do not affect
the patient in any salient manner. Furthermore, the reference relations of the
arguments deviate from the transitive prototype. This means, for example, that in
cases like ‘I fear him’ the nominatively marked Agent refers rather to a patient-like
participant that registers the effect of the event most directly, while the Patient
indicates a causer. Hence, it does not come as a surprise that numerous languages
reverse the marking relations in these kinds of case. Constructions like (977) reflect
the less transitive semantics of the events profiled more directly than typical
transitive clauses. The effects of inherent transitivity on the structure of clauses is
a very broad topic and we will not dwell on it here, since it only represents one part
358

of our study (for a graphic illustration of the topic in other languages, see Tsunoda
1985: 388 and Drossard 1991: 411f).
The lower degree of dynamicity also contributes to lower transitivity of
events. For example, liking or loving is not a dynamic action, but rather a state that
involves an experiencer and a stimulus. Dynamicity makes also otherwise a very
significant contribution to transitivity. States are in general marked less transitively
than dynamic events, as shown below, cf.

(982) à:ng-í shvm pé-ò-‘

3SG-AGT sword hang.on.shoulder-3.TR.N.PAST-N.PAST
‘He is putting on or wearing a sword’
(983) àng shvm dv-p ‘-shì-‘
3SG sword CAUS-hang.on.shoulder-REFL/MID-N.PAST
‘He is putting on a sword’
(984) àng shvm p ‘-shì-‘
3SG sword hang.on.shoulder-REFL/MID-N.PAST
‘He is wearing a sword’ (on-going state) (Dulong/Rawang, LaPolla 2000: 295)

(985) ham ii iyaad karat baanii

I.NOM this.ACC remember doing am
‘I am memorizing this’
(986) hamraa ii baat iyaad baa
I.DAT this.NOM remember is
‘I remember this’ (Bhojpuri, Verma 1990: 93)

(987) ishkiin nayilín y-o-ø-ø-keesh

boy girl her-PROG-he-CLASS-bother
‘The boy bothers the girl’
(988) ishkiin nayilín b-o-ø-ø-keesh
boy girl her-PROG-he-CLASS-bother
‘The boy bothers the girl’ (San Carlos Apache, Shayne 1982: 400f)

All the examples cited above illustrate how a single verb can appear in multiple case
frames based on dynamicity. As predicted, the employed construction is less
transitive in stative cases. In Dulong/Rawang, the differences between dynamic and
stative uses of the verb pé are very obvious and the shift from a dynamic to a stative
 359

use results in an intransitivization. Examples (982) and (983) denote transitive

events ((982) can also be interpreted statively), whereas (984) profiles a state. In
Bhojpuri, the verb ‘remember’ can appear in two different constructions according
to whether the event denoted is considered as dynamic or stative. The overall
transitivity of both events is rather low, but (985) is construed as more transitive
based on dynamicity. Examples from San Carlos Apache are crucially different from
the previous ones. Of these examples, (987) is the more dynamic one. The boy is
actively doing something in order to bother the girl. In (988), on the other hand, the
girl lets herself to be bothered by the boy who does not even have to be present, i.e.
thinking about the boy is what bothers the girl. What distinguishes between these
two events is consequently the active vs. non-active nature of the causer. (987) can
be regarded as more dynamic, since it involves an active causer that is lacking in
(988). In San Carlos Apache, the effects on structural transitivity are less drastic in
(982)-(986). There are no changes in the case marking.
Rather closely related to dynamicity is the concreteness of events, which also
makes a very significant contribution to linguistic transitivity. For example, dynamic
events are more concrete and salient than stative ones. Concrete events are encoded
differently from non-concrete (or irrealis) ones in a number of languages (the former
being more transitive in case a language has this kind of ‘split’). Non-concreteness
is here understood to cover all events the occurrence of which we have not directly
witnessed. This includes habitual, potential and desiderative events as well as events
that have yet not occurred (the future tense). All these events involve a potential
agent that is not carrying out a transitive action at the moment the clauses are
uttered, but is fully capable of doing so and may do so (the three last phases of
transitive events have not been realized yet). The basic semantics of events is the
same as in transitive clauses. For example, the event ‘I usually eat fried onion with
french fries’ is an instance of a non-concrete event in the view adopted here.
Differently from ‘I ate fried onion with french fries’, the agent is not involved in the
action profiled as we speak. This also implies that there is no concrete patient
participant directly impinged on. In what follows, we will examine the four different
non-concrete event types noted above. The last type differs from others in that the
given event is more likely to occur. In the other cases, the possibility of occurrence
is conditioned by different principles. We illustrate the constructions in the order
360

they are mentioned here. This correlates roughly with the probability of an actual
occurrence of events.
As potential are here regarded constructions involving (potential) agents
capable of performing the action denoted. In distinction from typical transitive
events, the agent is not carrying out a concrete action. Consequently, there is no
concrete patient present, either. A couple of examples of less transitive potential
constructions are the following, cf.

(989) k’o -i gišanc’q’uns sagan-s

man-NOM extract.I arrow-DAT
‘The man is pulling out the arrow’
(990) k’o-s gešaanc’q’e sagan-i
man-DAT extract.POT.I arrow-NOM
‘The man can pull out the arrow’ (Mingrelian, Harris 1991: 369)

(991) mam c ee potc kiyewwa

I that book read
‘I read that book’

(992) ma šc siõh clc kiycwann c puluwan

I.DAT Sinhala read.INF can
‘I can read Sinhala’ (Sinhala, Gair & Paolillo 1997: 32, 27)

(993) w Ç manu-na s§ pal-a

the man-ERG wood cut-PERF
‘The man cut (the wood)’
(994) w Ç manu(-na) na-e phu
the man(-ERG) eat-INF can-HAB
‘The man can eat’ (Newari, Givón 1985: 94)

Events profiled above are all basically transitive events. These clauses do not,
however, describe concrete events, but mere capabilities of human entities. The
human entity referred to is capable of performing the action at issue. The
constructions are fully neutral as regards the actual occurrence of events. The agent
may instigate an event or may choose not to do so.
 361

The second type is illustrated by desiderative constructions. As the label

implies, these refer to events an agent would like to partake in. As regards the
probability of an actual occurrence of events, desiderative constructions are very
similar to (989)-(994). The distinction between the two types is based primarily on
the semantic differences between these and typical transitive constructions. Potential
constructions describe a mere capability of an agent without any reference to
whether the event in question will ever take place. In desiderative constructions, on
the other hand, the agent would like to partake in an event, but may be prevented
from doing so (perhaps for reasons beyond his/her control). Potentiality is usually
(but not necessarily) an implicit part of wanting to do something, since it implies
that the agent has the knowledge needed in order to perform the action in question.
A couple of desiderative constructions are illustrated below, cf.

(995) bima-n-ra xenan-Ø koko-ai

PN-ERG-AS guava-ABS suck-INC
‘Bima is eating guava (fruit)’
(996) Bima-Ø-ra xenan-Ø koko-kas-ai
PN-ABS-AS guava-ABS suck-DES-INC
‘Bima wants to eat guava’ (Shipibo-Conibo, Valenzuela 1997: 196)

(997) amma kuttiye atik’k’anam

mother.NOM child-ACC beat-anam
‘The mother must beat the child’
(998) ammak’k’c kuttiye atik’k’anam
mother-DAT child-ACC beat-anam
‘The mother wants to beat the child’ (Malayalam, Mohanan & Mohanan 1990: 45)

In Shipibo-Conibo and Malayalam (cf. also (916)-(917) from Pitta-Pitta),

desiderative constructions rank lower in transitivity than typical transitive ones. In
Shipibo-Conibo, the difference between concrete and desiderative events is
expressed by omitting the ergative marking of the Agent. This is rather natural, since
we are not dealing with a concrete agent. In Malayalam, the Agent of desiderative
clauses appears in the dative that is the case of experiencer subjects in the language.
The next type undet study is illustrated by constructions denoting habitual
events. Habituativity is rather generally related to lower degrees of transitivity (cf.
362

e.g. Siewierska 1984: 254, Moreno 1985: 177, Lindvall 1998: 160f, Lazard 1998:
210). Habituativity is, for example, related to imperfective aspect, as noted by
Lindvall (1998: 119). Habitual constructions share many common features with
constructions involving indefinite Patients (cf. e.g. (375)-(376) from Turkish). Both
involve a patient whose precise identity is irrelevant. Habitual constructions could
even be regarded as a subtype of indefinite Patient constructions. Structurally, these
constructions are identical in many languages (for example, antipassives are used
to express both in numerous languages). In the present study, these two
constructions are, though, regarded as distinct. This is primarily due to the fact we
aim at showing what kinds of semantic differences are sensitive to transitivity of
clauses. Despite the obvious similarities, these two constructions also differ from
each other, which makes it necessary to distinguish between them. The distinction
is here based on whether the given event has occurred and whether we are aware of
this. As habitual are classified only constructions that profile events for the actual
occurrence of which we do not have any direct evidence. Hence, habitual
constructions do not refer to any specific event. The ‘indefiniteness’ of the event
itself results in the indefiniteness of the patient. In the case of indefinite patient
constructions, on the other hand, the indefiniteness of the patient is a property of the
patient and not of the event. The event profiled may have occurred. Differently from
the potential and desiderative constructions, habitual constructions denote frequently
occurring events that can be regarded as typical of the entities referred to.
Consequently, it is rather likely that the event in question will take place again in a
rather immediate future. What potential constructions imply is simply that the entity
referred to is capable of the denoted action. A couple habitual constructions are
exemplified below, cf.

(999) àng-í pé tiq-chvng zvt-ò-‘

3SG-AGT basket one-CL weave-3.TR.N.PAST-N.PAST
‘He is weaving the basket’
(1000) àng pé zvt-‘
3SG basket weave-N.PAST
‘He weaves baskets’ (Dulong/Rawang, LaPolla 2000: 285)
 363

(1001) ibayu nandé amak

PASS.weave mother mat
‘Mother wove a mat’
(1002) nandé mbayu amak
mother ACT.weave mat
‘Mother is weaving a mat/mother weaves mats (as an occupation)’ (Karo Batak, W oollams
1996: 214)

(1003) no-laha te doe

3R-search CORE money
‘He is looking for money’
(1004) no-me-laha (*te doe)
3R-FREQ-search
‘He is always looking’ (Tukang Besi, Donohue 1999: 272)

(1005) e vuke-i ~ã
3SG help-i me
‘He helped me’ (on a specific occasion)
(1006) e d ~ã v ‘§-vuke
3SG HABITUAL v Ԥ-help
‘He helps (habitual)’ (Fijian, Schütz 1985: 208)

(1007) sobaka kusajet po tal'ona

dog.NOM bite postman
‘The dog bites the postman’
(1008) sobaka kusajet-sja
dog.NOM bite-REFL
‘The dog bites’ (Russian, Comrie 1985: 319)

The two first constructions illustrate cases in which the target is explicitly referred
to. Hence, the nature of the patient is relevant for the right interpretation of the
construction. For example, in (1002) from Karo Batak, the habituativity is restricted
to weaving mats and not to weaving in general. In the last three cases, the Patient
has been omitted altogether. These cases can be regarded as more general, since the
identity of the patient is completely irrelevant. What is interesting in the examples
from Karo Batak is that a construction typically (cross-linguistically) regarded as
less transitive is used in the description of (completed) concrete events. In all other
364

cases, habituativity results in an obvious transitivity decrease.

The last type comprises future events. They are very likely to take place, but
have not occurred yet. What primarily distinguishes these from concrete transitive
events is that we cannot yet know whether the event will be successful or not. The
most typical examples of the less transitive marking of future events are provided
by tense-conditioned splits attested in many predominantly ergative languages. The
differences between future and other events coincides with the concrete/non-
concrete distinction noted above. The former are similar to non-concrete events,
whereas events in the present and (remote) past tenses are more similar to concrete
events (even if the event is not taking place in the latter case). As opposed to cases
illustrated thus far, future events are distinguished from concrete ones only on the
basis of tense. In the three previous cases, the non-concreteness has been motivated
by other features and the actual occurrence of events has been less likely than in the
following cases. Below, we will illustrate different tense/aspect splits and discuss
their relevance for transitivity. We begin by illustrating splits conditioned by tense,
since (as noted above) they also, in a way, illustrate the concrete vs. non-concrete
distinction.
It is common for all the languages with a tense-conditioned split of some kind
that the future tense follows a kind of nominative-accusative pattern, whereas the
past tense (and perfective aspect) employs an absolutive-ergative pattern (which in
many languages means that the Agent is marked ergatively in the past (and present)
tenses, whereas the marking is omitted in the future). There is obvious variation in
the present. In some languages, clauses in the present tense are nominative-
accusative, while in others the marking is absolutive-ergative. Tense splits are
illustrated and discussed in the following, cf.

(1009) raam-ne ravii-ko piitaa

Ram-ERG Ravi-ACC beat-PERF
‘Ram beat Ravi’
(1010) raam ravii-ko piittaa hai
Ram.NOM Ravi-ACC beat-IMPERF be-PRES
‘Ram beats Ravi’ (Hindi, M ohanan 1994: 70)
 365

(1011) pirala patya-na õayi mantu-yi

yesterday eat-PAST I meat-ACC/DAT
‘Yesterday I ate the meat’
(1012) patakari õayi patya-Ru mantu
tomorrow I eat-FUT meat
‘Tomorrow I will eat the meat’ (Gariera, Blake 1977: 18)

(1013) wu gún-d c g ]-c walõá-

boy-ERG hit-PAST girl-ACC
‘The boy hit the girl’
(1014) yargúl-u õg c g ] - u õ m ]g
woman-ERG hit-PRES man
‘The woman is hitting the man’ (Mbabaram, Dixon 1991: 367ff)

(1015) tomb c-n c (tha õgon-n c) s cjik phalli

Tomba-ERG (sickle-ERG) grass cut
‘Tomba cut the grass (with a sickle)’
(1016)tomb c-gi kophi th ck-e
Tomba-GEN coffee drink-PERF
‘Tomba has drunk coffee’ (Manipuri, Bhat & Ningomba 1997: 109, 130)

The split of Hindi illustrates a typical tense/aspect-conditioned ergative vs.

nominative split. The Agent appears in the ergative in the perfective aspect of the
past tense. The marking of the Patient is insensitive to tense. In Gariera, on the other
hand, an explicit accusative/dative marking of the Patient is restricted to the past
tense. This can also be explained by referring to the concrete vs. non-concrete
distinction. The Patient in (1012) refers only to a possible patient that is, however,
a rather likely target of a transitive action. Example (1011) profiles an event that has
already occurred and caused a salient change-of-state in the patient. What makes the
examples from Gariera typologically interesting, is that we are not dealing with split
ergativity, but rather split accusativity, since the marking of Patient is conditioned
by tense. The split attested in Mbabaram is similar to that in Gariera, since Patients
are marked only in the past tense. The ergative marking of Agent is neutral in this
respect. The status of this split is, however, rather vague, since this kind of split
marking is attested in only four examples cited by Dixon. However, it may be
justified to label this as a tense-conditioned split, since the accusative marking is
366

exclusively restricted to the past tense. The examples from Manipuri serve primarily
as a curiosity, since the split is not directly conditioned by concreteness of events.
According to Bhat & Ningomba (1997: 130), the ergative may be replaced by the
genitive, which is most frequent, if the verb is the present perfective form.
Above, we adopted the view that tense-conditioned splits are best explained
by differences in the concreteness of events. The other (and more generally
accepted) view is the so-called viewpoint- or pivot-theory, adopted, for example, by
DeLancey (1982) and Dixon (1994: 97ff). The leading principle of these theories is
that the unmarked argument of transitive clauses coincides with the viewpoint of the
clause. Because this argument is marked identically to the sole argument of
intransitive clauses (that, as the only argument present, is the viewpoint), it is
claimed that it is the viewpoint of the clause. The high grammatical status of either
argument also reflects the high topicality of the participant referred to. This means
that the agent is the prominent participant in the future (and present) tense, while the
focus shifts in the past tense. The agent really is more important in the future tense,
since it depends primarily on the agent whether an event will occur or not. On the
other hand, patient is the only relevant participant, when the agent ceases to act. We
can observe the event only through the resultant state of the patient that is the only
salient participant (cf. above). Even if there are valid arguments for adopting the
viewpoint theory, we emphasize the concreteness principle in the present study,
simply because this is more appropriate for our purposes. In this view, explicit
argument marking is sensitive to semantic roles and only arguments referring to
concrete participants are marked explicitly. Furthermore, viewpoint theory is not
applicable to splits attested in Hindi, Gariera or Mbabaram. In Gariera, the Patient
is clearly the less pivotal argument in the past tense. In Hindi, on the other hand,
there is no pivotal (i.e. zero marked) argument in (1011). We do not, however, wish
to reject the viewpoint theory altogether, but regard it as an alternative to the
analysis proposed here.
In cases illustrated so far, the default interpretation has been that we have
some evidence that the event will be successfully completed (this may also be
irrelevant for the marking). In some languages, highly transitive constructions imply
that events have occurred and that the speaker is aware of that. Thus, clauses in the
possibilitive mood (cf. Kibrik 1994: 353) illustrate deviations from the transitive
 367

archetype. This seems to be typical of some Caucasian languages, cf.

(1017) st’alin-ma tav-is-i mt’r-eb-i ga-(Ø-)žlit’-a

Stalin-ERG his own-NOM enemy-PL-NOM he.has.exterminated.them
‘Stalin exterminated his enemies’
(1018) st’alin-s tav-is-i mt’r-eb-i ga-(Ø-)u-žlet’-i-a
Stalin-DAT his own-NOM enemy-PL-NOM he.has.exterminated.them.?
‘Stalin has apparently exterminated his enemies’ (Georgian, Hewitt 1995: 203)

(1019) ko i-k qvilups / doqvilu pe ¯i

man-ERG he.kill.it.I / he.kill.it.II pig.NOM
‘The man kills/killed a pig’
(1020) ko i-s uqvilun pe ¯i
man-DAT he.kill.it.INV.III pig.NOM
‘The man has killed a pig’ (Laz, Harris 1985: 52, 287)

(1021) gabirr-inh/-nda nganhi gunda-y

girl-ERG 1SG.ACC hit-PAST
‘The girl hit me’ (just now, recently, I still got a mark to prove it)
(1022) gabiirr-ngun nganhi gunda-y
girl-ERG 1SG.ACC hit-PAST
‘The girl hit me’ (some time ago, neutral sense) (Guugu Yimidhirr, Haviland 1979: 51)

Examples from Georgian and Laz (see also Kibrik (1994: 353) for Archi) illustrate
what is labelled as inverse in Caucasiology (see the glosses in (1020)). In these
cases, possibly occurred events are encoded differently from concrete ones. The
examples from Guugu Yimidhirr are crucially different from (1017)-(1020). First
of all, the transitivity is not decreased in any salient way. (1022), can, however, be
regarded as somewhat less transitive than (1021), since only in the first case we have
direct evidence for the occurrence of the profiled event. Semantically, (1022) is
consequently similar to (1018) and (1020). The basic transitive marking is related
to ‘witnessed concreteness’, which means that the transitive marking appears only
if the given event has actually occurred and we have some kind of direct evidence
for this.
Examples thus far illustrate transitive vs. less transitive marking of clauses
conditioned by the concreteness of events. The mere concrete and directly
368

observable occurrence of events is, however, not always a sufficient criterion for
high transitivity, but events have to be successfully completed in order to be marked
as highly transitive. The patient retains its role (cf. Croft 1994: 96 ‘[...] a participant
that is objectively considered to be a patient...’), but is affected to a lesser extent.
Only successfully completed events are marked transitively. This means that events
in the perfective aspect are more transitive than those in the imperfective aspect
(understood in a semantic sense). A couple of examples of this are the following,
cf.

(1023) ramesh-e pen kh crid-y-i

Ramesh-ERG pen.FEM buy-PERF-FEM
‘Ramesh bought the pen’
(1024) ramesh pen kh crid-t-o h c-t-o
Ramesh pen.FEM buy-IMPF-MASC AUX-IMPF-MASC
‘Ramesh was buying a pen’ (Gujarati, DeLancey 1981: 628f, cited from Mistry 1976: 257,
245)

(1025) s~ faitau (‘uma) e ulika le tusi

PAST read (all) ERG Ulika ART letter
‘Ulika read the (whole) letter’
(1026) s~ faitau ulika i l=a=na tusi
PAST read Ulika LD ART=POSS=3SG letter
‘Ulika read her letter’ (lit. Ulika read in her letter) (Samoan, Mosel & Hovdhaugen 1992: 111)

(1027) der mann bau-te ein haus

ART.NOM man build-3SG.PAST ART.ACC house
‘The man built a house’
(1028) der mann bau-te an ein-em haus
ART.NOM man build-3SG.PAST PREP ART-DAT house
‘The man was building a house’ (German)

(1029) ko e kai e mautolu a talo

PRES eat ERG 1.PL.EXCL ABS taro
‘W e’re eating up taros’
(1030)ko e kai a mautolu he talo
PRES eat ABS 1.PL.EXCL at taro
‘W e’re eating (some) taro’ (Niuean, Seiter 1980: 34)
 369

Above, the first example of each pair illustrates a successfully completed event.
Hence, the marking is transitive. In the second cases, the completion is less
successful or total. These differences are rather natural (and frequently attested),
since imperfective aspect is also related to a lower degree of affectedness. Examples
from Niuean differ from the other examples in that they are in the present tense. The
differential marking is due to differences in the expected degree of completion.
Examples above illustrate cases in which the lower degree of affectedness is
very closely related to a less thorough completion of events. This is due to the fact
that the examples (with the possible exception of Gujarati) present cases in which
the patient can be regarded as an instance of ‘incremental theme’ in the spirit of
Dowty (1991). In (1023)-(1030), the degree of completion of an event coincides
with the amount of affectedness. For example, painting a house consists of
numerous individual instances of painting. The patient targeted by an individual
subevent is usually completely affected. The agent is typically capable of controlling
the number of subevents. If a full affectedness involves, say 200 instances of
painting, an event that involves only 100 of them is construed as involving a less
affected patient and the event as a whole is construed as less thoroughly completed.
The changes in the affectedness parameter do not follow from lower degrees of
affectedness of individual instances, but the crucial factor is the number of affected
‘sub-patients’ and completed ‘sub-events’. The extent of the affected part is smaller
than expected (or possible). The less thorough completion of events is based on this.
Also the intimate relation of the nature of transitive events and the affected
participant makes a contribution here. There are, however, also cases in which a
change in the affectedness does not directly correlate with a less than perfect
completion of the event. A couple of examples are illustrated below, cf.

(1031) ha-panek si juan este i lalahi

ERG 3SG-beat UNM Juan this the men
‘John beat these men’
(1032) amanek si juan nu este i lalahi
AP-beat UNM Juan OBL this the men
‘John pounded on these men’ (Chamorro, Cooreman 1988: 582)
370

(1033) wac oi jaci L o'awo

brother-A/ERG sister-P/NOM beat
‘Brother beat sister’
(1034) wac oi jacu ’i buRo
brother-A/ERG sister-C.POSS hit
‘Brother hit sister’ (Tindi, Kibrik 1985: 300)

(1035) ti he-v
tree chop-PRES
‘Chop down a tree’
(1036) ti-m he-the-v
tree-DAT chop-INTR-PRES
‘Chop on a tree’ (W aris, Foley 1986: 109)

All the examples presented above illustrate successfully completed events that do
not affect the patient completely. The event as a whole is successfully completed,
but differently from typical transitive events (denoted in the first example of each
case), the patient is not completely affected. As opposed to (1023)-(1030), the lower
degree of affectedness does not follow from a smaller number of completed sub-
events. The action simply did not affect the patient as a whole as totally as predicted.
For example, the act of beating (differently from a mere instance of hitting) implies
complete affectedness. This is the case in (1031) and (1033), but not in (1032) and
(1034). In Chamorro, the structure employed is antipassive, whereas in Tindi the
Patient appears in the possessive, if the patient is not completely affected. Notice
that example (1034) has to be regarded as a rather marginal example of a true
transitivity alternation, since the lexical verb is different, which determines the
marking. The conditioning factor is, however, very similar to that in Chamorro.
Examples from Waris illustrate the construal of different instances of ‘hitting’.
The other case in which a lower degree of affectedness is not in any direct way
related to a less perfect completion of events is provided by the more explicit
marking of animate patients in comparison with inanimate ones. This usually
follows from the inherent definiteness of animate patients. However, the fact that
animate patients experience the effect of transitive events more intensively than
inanimate also makes a contribution. For example, an act of hitting affects a human
patient more directly than an inanimate one. An example of an alternation in which
 371

this can be said to be the primary conditioning factor is given below95, cf.

(1037) Ilyatjari-lu pony tati-nu

Ilyatjari-ERG pony climb-PAST
‘I. mounted the pony’
(1038) nyantju wala winki puli tati-nu
horse very quickly hill climb-PAST
‘The horse climbed the hill flat out’ (Pitjantjatjara, Rose 1996: 295)

Example (1037) describes a transitive event that directly affects the patient, while
(1038) profiles a process that is extended to another participant which is, however,
not affected by it in any direct way (see Rose: ibidem). The basic event itself is the
same, but the animacy of the patient distinguishes between the two events. The use
of the transitive construction is restricted to events involving animate patients.
In all the cases illustrated above, the agent is successfully participating in a
transitive event that, however, is not completed in the way expected based on the
verb semantics. The other type of less completed events is illustrated by cases in
which the event has failed to occur, cf.

(1039) njuntulu-lu ø-npa-tju pantu-nu õatju

2SG-ERG ø-2SG-1SG spear-PAST 1SG.ABS
‘You speared me’
(1040) njuntulu-lu ø-npa-tju-la pantu-nu õatju-ku
2SG-ERG ø-2SG-1SG-la spear-PAST 1SG-DAT
‘You speared at me’ ‘You tried to spear me’ (W arlpiri, Hale 1973: 336)

(1041) lilmi-i guñ g-Ø-a-n-na nda giimbu

man-ERG watch Ø-he-do-PAST-it CONJ kangaroo
‘The man saw the kangaroo’
(1042) lilmi guñ g-Ø-n-nda giimbu-yi
man watch Ø-he-do-PAST kangaroo-GEN
‘The man was watching for kangaroos’(Alawa, Sharpe 1976: 512)

95
In many cases, the definiteness and the more intensive experiencing of the effect of event overlap
and it is not possible to distinguish between them.
372

As opposed to examples cited above, events denoted in (1040) and (1042) have fully
failed to occur, which implies that the patient has not been affected. In both cases,
an agent is intending to carry out a transitive action, but fails to initiate one.
Examples above exemplify what is generally (see e.g. Guerssel et al 1985: 50 and
Levin 1993: 41) labelled as conative alternations. Conative alternations describe
attempted actions instead of actual ones. The lower degree of transitivity in (1040)
and (1042) in comparison with (1039) and (1041) is very obvious, since there is no
affected patient, even if an action is targeted at a patient. We could even claim that
these events are fully intransitive, since they involve one central participant only.
However, it is necessary to distinguish these kinds of event from completely
intransitive ones, since the patient is present in the cognitive structure of the events.
This is not the case in genuinely intransitive events like ‘he runs’. The less transitive
nature of (1040) and (1042) is genuinely conditioned by the lower transitivity of
events denoted. In (1009)-(1016), for example, the less transitively marked events
are, despite their marking, possibly transitive events that may be successfully
completed.
Examples presented thus far underline the significance of high affectedness
for linguistic transitivity. Also the non-volitionality or inactivity of patients makes
a contribution to this. Typical patients are passive participants that simply undergo
the effect of actions without being able to control this in any way (see e.g. Givón
1995: 76). Thus, we could assume that there are languages in which ‘active’ patients
are marked differently from (and less transitively than) typical patients. This is
exactly what we found in languages that have two causatives used according to the
degree of volitionality associated with the causer, cf (see also (1207)-(1218) below):

(1043) en köhogtettem a gyerek-et

I caused.to.cough the child-ACC
‘I made the child cough’ (e.g. by slapping him/her on the back)
(1044) en köhogtettem a gyerek-kel
I caused.to.cough the child-INSTR
‘I got the child to cough’ (e.g., by asking him/her to do so) (Hungarian, Payne 1997: 186)

(1045) pia ga anssi o konsaato e ik-ase-ta

PN NOM PN ACC concert to go-CAUS-PAST
 373

‘Pia made Anssi go to a concert’

(1046) pia ga anssi ni baa e ik-ase-ta
PN NOM PN DAT bar to go-CAUS-PAST
‘Pia let Anssi go to a bar’ (Japanese)

(1047) murhamies tappo-i hän-et

murderer.NOM kill.PAST-3SG s/he-ACC 1
‘The murderer killed him/her’
(1048) mies anto-i hän-en kuolla
man.NOM let.PAST-3SG s/he-ACC 2 die
‘The man let him die’ (Finnish)

The causee is less volitional and active in the first example of each pair. The basic
transitive pattern is employed in these cases, while volitional causees are referred
to by less transitive constructions. In (1043), the causee is not acting volitionally,
since the coughing follows from a physical action that usually causes coughing. In
(1044), the same action follows from a request, and there is no physical
manipulation involved. The causee could also refuse to act in the desired way in
(1044). The examples from Hungarian illustrate differences between manipulative
(physical) and directive (verbal) causation (cf. Song 2001: 276f). Examples (1045)
and (1046) profile events that involve verbal manipulation only. These examples are
distinguished from each other on the basis of the degree of volitionality related to
the causee. In (1045), the degree of volitionality is lower and the structure employed
is typically transitive. In (1047) from Finnish, the patient is also clearly a passive
and non-volitional participant and the Patient appears in the so-called t-accusative.
Example (1048), on the other hand, is more appropriate, if the patient wished to die,
e.g. because s/he was suffering immensely and wanted to end the misery. In this
case, the Patient appears in the n-accusative that is diachronically related to the
dative.96 The use of the pattern usually implies also a less direct causation, but this
does not have to be the case. We can use the clause also in case a doctor
(purposefully) gives a lethal dose of morphine to a patient who wishes to die, in
which case the causation per se is very direct. The patient cannot affect the change-

96
I thank Nobufumi Inaba for this remark.
374

of-state in any way, but its participation in the event can be regarded as volitional.
Examples from Finnish are rather marginal, since the variation exemplified above
is restricted to personal pronouns97. Examples above could also be said to illustrate
differences between complete and less complete transitivizations. The claim that the
kinds of difference illustrated above are due to changes in volitionality is justified,
since the change-of-state in the patient is the same in all cases. Only the degree of
volitionality/activity associated with the patient varies.
The other kind of transitivity alternation due to changes in individual
transitivity features is naturally illustrated by alternations related to changes in the
agentive features of events. As stated above, a typical transitive event involves a
volitionally acting, human agent that controls the event and is not affected by the
event in any direct way. The notion of agency is very relevant for high transitivity,
and any deviation from this prototype may result in a transitivity alternation (see e.g.
Kegl 1985: 135). As also noted above, changes in agency parameter are completely
independent of affectedness parameter, which means that the agency can vary
without this having any consequences for the affectedness of the patient. Below, we
illustrate agency-related alternations. Each relevant parameter associated with high
agency is illustrated and discussed in turn.
Decreases in agency can follow from the inherent nature of events, the
inherent nature of the instigator, or the lower degree of agency may be situational
(i.e. due to neither of the previous reasons). Typical examples of cases in which the
low degree of agency involved inheres in the nature of events are provided by
experiencer constructions. We can control our feelings or perception only to a
limited extent. Hence, the agency associated with these events is inherently low, cf.

(1049) mha ne jodhpur cokho l~ge

me ne Jodhpur nice seems
‘I like Jodhpur’ (Maithili, Magier 1990: 215)

(1050) õa-s debs brlags-so õ

I-ERG book lose-PERF.INVOL

97
excluding se/ne (‘it’/’they’) that have originally been used to refer to inanimate entities, but which
in colloquial Finnish refer to humans as well.
 375

‘I lost a/the book (involuntarily)’

(1051) õa-la debs rñed-byu õ
I-DAT book find-PERF.INVOL
‘I found a/the book’ (Lhasa Tibetan, DeLancey 1984: 136)

(1049) illustrates an experiencer construction typical of many South Asian

languages. The case particle ne marks Patients as well as experiencer subjects
(Magier: ibidem). In Lhasa Tibetan, the only difference to a typical transitive
construction is the ‘involitional verb morphology’. This is semantically rather
consequent, since ‘losing’ is usually not instigated with intent. This is even more
obvious in the case of finding, as stated by DeLancey (ibidem) and shown in (1051).
The reduced degree of agency inheres in the semantics of events in (1049)-
(1051). The precise nature of the instigator is irrelevant, even if the participation in
experiencer constructions implies that the ‘agent’ be animate. Animacy is in general
the most prominent feature of agency. Typical agents are conscious, animate entities
(typically humans). Only animate entities are capable of volitionally instigating
events they can control. Inanimate entities are inherently less typical instigators of
events. They cannot instigate events intentionally nor can they consciously target
their action at patients. This is directly reflected in the clause structure in a number
of languages, cf.

(1052) adam-e ji¨“¨ alebt’c

man-ERG bridge.III.ABS III.destroy.PAST
‘The man destroyed the bridge’
(1053) dama-n ji¨“¨ alebt’c
river-ERG bridge.III.ABS III.destroy.PAST
‘The river destroyed the bridge’ (Tsakhur, Schulze 1997: 58)

(1054)I tûûlî pwaxi eli a kaavo

3SG dry child that ERG Kaavo
‘Kaavo is drying the child’
(1055)(I) thâlî daan ru ciiy-ena
(3SG) block road ERG tree-this
376

‘This tree has blocked the road’ (Nêlêmwa, Bril 1994) 98

(1056) joe lán ©hwe

Joe die.TRANS
‘S/he killed Joe’
(1057) ‘éyaki k’é lán ©we
disease k’é died
‘S/he died from sickness’ (Hare, DeLancey 1984b: 186, 189) 99

(1058) lam cya wælikandak hæduwa

child.NOM sand-hill.INDEF make.PAST
‘The child makes a sandpile’
(1059) hulange õ wælikandak hæduna
wind.INSTR sand-hill.INDEF make.P.PAST
‘A sandpile formed (because of the wind)’ (Sinhala, Gair 1990: 16)

The latter example of each pair denotes an event instigated by an inherently non-
agentive entity. The event itself allows both readings, i.e. the basic nature of the
event remains more or less the same. Hence, the relevant factor is the decreased
agency. Tsakhur and Nêlêmwa exemplify languages with two ergative case markers
whose use is conditioned by animacy. In Tsakhur, this is the only difference between
the two clauses. Animate Agents are marked by -e, whereas -n is attached to
inanimate Agents. Hence, the structural transitivity of the clauses is in principle the
same.100 Examples from Nêlêmwa are very similar in this respect, since also in
Nêlêmwa a distinct ergative case marking distinguishes between typical and less
typical agents. Differently from Tsakhur, this change is accompanied by an optional
neutralization of the verb agreement. Therefore, (1055) can be regarded as less
transitive than (1054) in which the verb obligatorily agrees with the Agent.
Examples from Hare and Sinhala illustrate more obvious transitivity decreases. In

98
Examples cited by Ross Clark in his guest lecture held in Frank Lichtenberk’s class ‘Grammatical
patterns in Oceanic languages’ at LSA 2001.

99
Some inanimate instigators, like ‘liquor’ can appear in the typical transitive scheme, but many
names of sicknesses cannot, DeLancey (ibidem).

100
It is not discussed by Schulze, whether the differences in the marking have other consequences that
would be reflected somehow indirectly.
 377

Hare, inanimate Agents are marked by a specific particle. Moreover, the verb is
morphologically intransitive in (1057). In Sinhala, inanimate Agents appear in the
instrumental case, which is accompanied by morphological changes in the verb. In
(1059), the instrumentally marked Agent is freely omissible. The meaning shifts to
‘the sandpile formed’.
Examples (1052)-(1059) illustrate an obligatorily differential marking of
clauses with inanimate Agents. Moreover, inanimate instigators can be only
optionally encoded differently from typical ones. In this respect, they differ from
typical transitive events that obligatorily appear in highly transitive frames. A couple
of examples are illustrated below, cf.

(1060) q’u ¨t’i-li lo e ¨wq’ni

thunder-ERG boy(III).ABS III.frighten.AOR
‘The thunder frightened the boy’
(1061) q’u ¨t’i-li-tN|:’i• lo e ¨wq’ni
thunder-SAF-SUB.ABL boy(III)ABS III.frighten.AOR
‘The boy was afraid of the thunder’ (Archi, Schulze 1997: 58)

(1062) na tapuni e le matagi le faitoto’a

PAST close ERG ART wind ART door
‘The wind closed the door’
(1063) na tapuni i le matagi le faitoto’a
PAST close LOC ART wind ART door
‘The wind closed the door’ (Samoan, Mosel & Hovdhaugen 1992: 423ff)

(1064) kuume tappo-i miehe-n

fever.NOM kill-3SG.PAST man-ACC
‘The fever killed the man’
(1065) mies kuol-i kuume-eseen (*karh-un)
man.NOM die-3SG.PAST fever-ILL (bear-ILL)
‘The man died of fever (*of a bear)’ (Finnish)

(1066) veter-Ø razbi-l-Ø okn-o

wind-NOM.SG.MASC break-PAST-SG.MASC window-ACC
‘The wind broke the window’
(1067) okn-o razbi-l-o vetr-om
window-ACC.SG.NEUTR break-PAST-SG.NEUTR wind-INSTR
378

‘The wind broke the window’ (Russian, examples courtesy of Katja Gruzdeva)

Archi and Samoan are very similar to Tsakhur, since the only differences between
clauses involving typical and untypical Agents is the marking of Agent. In Archi,
inanimate Agents can be marked both in the ergative and in the subablative. In
Samoan, the variation is between ergative and locative cases, which is exclusively
restricted to cases like (1063) (as explicitly stated by Mosel & Hovdhaugen). Also
in Finnish and Russian, the patterns illustrated in (1065) and (1067) are restricted
to the description of events involving untypical instigators.101 In Finnish and
Russian, the reduced agency results in a more massive morphosyntactic
manipulation of the clause structure.
Examples (1052)-(1067) illustrate the differential marking of inanimate
instigators that can be regarded as ‘force’ (cf. e.g. Van Valin & LaPolla 1997: 85).
Typical instances of this category are, for example, forces of nature. Even if events
instigated by these entities are obviously less transitive, many languages encode
them using a typical transitive construction. Archi, Samoan, Finnish and Russian are
also examples of this, as shown above. Even if many languages allow forces to be
expressed as Agents, they are usually more ‘conservative’ in the encoding of
instruments as typical Agents (cf. DeLancey 1984b: 186), as illustrated below, cf.

(1068) li-ji kuu-õku õai õantamaji tiinta

3SG-ERG rain-ERG me find in.the.middle
‘The rain caught me in the open’
(1069) ntia-ku tuar ntati-nti-ji
stone-ERG snake crush-nti-PAST
‘The stone crushed the snake’ (Kalkatungu, Blake 1979: 32, 89)

(1070) õity-iya-kanta kuralata pulmpara

fire-ERG-TR.PAST scatter.IND grasshopper.ABS
‘The fire scattered the grasshoppers’
(1071) palata-lkanta õawuwa tu õal-ulu-ya
hit.IND-they.TR.PAST dog.ABS stick-COM -ERG
‘The stick hit the dog’ (lit. ‘they having a stick, hit the dog’) (Yukulta, Keen 1983: 248)

101
These kinds of alternation can also be regarded as indirect, see 5.3.2. for a more detailed analysis.
 379

(1072) kotúé’ ye-wéhx ©

liquor 3OBJ-killed
‘Liquor killed him/her’
(1073) *gof© yejai tá’en ©se
axe glass broke.TR
(The axe broke the glass)
(1074) gof© k’é yejai tá’©to
axe k’é glass broke.INTR
‘The window got broken by an axe’ (Hare, DeLancey 1984b: 186f)

(1075) havaa-ne patte bik her dyie the

wind-ERG leaves.NOM scatter give.PERF be.PAST
‘The wind had scattered the leaves’
(1076) ?patt har-ne šiišaa tod diyaa
stone-ERG glass.NOM break give.PERF
‘?The stone/rock broke the glass’ (Hindi, Mohanan 1994: 75)

(1077) rwàay p-háan tráak

tiger CAUS-kill buffalo
‘The tiger killed the buffalo’
(1078) *mìit p-háan tráak
knife CAUS-kill buffalo
(The knife killed the buffalo) (Kammu, Svantesson 1983: 104)

(1079) iskà taa buudè koofàa

wind.FEM 3SG.F.PERF open.IV door
‘The wind opened the door’

(1080) *wukaa taa yankà shaanuu

knife.FEM 3SG.F.PERF slaughter.I cow.PL
(The knife slaughtered the cows) (Hausa, Heide 1989: 61)

Kalkatungu and Yukulta allow, in principle, all kinds of instigator in the Agent slot
(that is characterized by the ergative case marker). However, certain modifications
are necessary, if the Agent is an instrument. In Kalkatungu, the Agent appears in the
ergative despite the obvious changes in the semantic role. What distinguishes (1068)
and (1069) is the verb morphology. For clauses like (1069) to be grammatical, the
affix -nti- has to be attached to the verb. In Yukulta, the ergative case marker cannot
380

be attached to instruments (as Agents) alone, but the ergative has to be accompanied
by the comitative, as shown in (1071). In Hare, the verb has to be intransitivized in
order that an instrument can occupy the (untypical) Agent slot. In some cases,
instruments can appear in ‘k’é constructions’ as can less typical instigators. Also
here, the instrument is understood as being manipulated by an agent. Hindi, Hausa
and Kammu all illustrate languages in which instruments in the Agent slot are either
less acceptable than other kinds of instigator (Hindi) or they are completely
disregarded as Agents (Hausa and Kammu). The differential marking of forces and
instruments follows, since, even being incapable of instigating events with intent,
forces can be primarily responsible for the occurrence of events. Instruments, on the
other hand, are very typically manipulated by humans, which makes it impossible
for them to be primary causes of events (cf. also (401) from Amharic). As might be
predicted, instruments are disallowed as Agents in ergative languages, since the role
of agent is morphologically focussed on. In languages like Kammu and Hausa, the
ungrammaticality is less obvious. In Kammu, the ungrammaticality can perhaps be
explained by referring to the causative marking of the verb. As illustrated in (741)-
(743), morphological causation implies direct causation, which makes (1078)
ungrammatical.
Above, we have illustrated the less transitive marking of constructions
denoting events, the inherent nature of themselves or that of their instigators
excludes high agency. Languages vary in whether they allow less typical instigators
to appear in the typical transitive frame or not. Finnish, Russian, Archi and Samoan
are examples of the latter kind of language, while Tsakhur, Sasak and Sinhala
exemplify the latter language type. Hare exemplifies both types. It was shown that
instruments are the least typical agents of all. Many languages that allow forces in
the Agent slot, disallow instruments. This is what we should expect, since forces
have one agentive feature more, since they can alone instigate transitive events,
which is not possible for instruments. Furthermore, they are not under external
control by an agent (DeLancey 1984b: 181). Otherwise, forces are very untypical
agents. Humans along with other higher animates differ crucially from inanimate
entities in this respect. As a result, unlike forces or instruments, human agents have
the potentiality of appearing in the transitive frame (cf. Duranti & Ochs 1990: 12).
They are always potential agents the agency of which can, however, vary drastically.
 381

The potential agency also increases the number of different alternations applicable
to clauses with human Agents. A change in each of the relevant features can result
in a transitivity alternation. Forces, on the other hand, always instigate events
involitionally and they are incapable of control. Hence, features like volitionality
and control are irrelevant. Below, we will illustrate further examples of alternations
conditioned by agency. The presentation proceeds also here from mild to more
severe changes. It has to be stressed that, even if we present the illustrated types as
distinct, there are clear overlaps. Many languages may express many (even all?) of
the alternations under study by a single alternation type. Since the goal of the
following presentation is to show in detail what can result in a transitivity
alternation, possible (minor) overlaps have to be accepted.
One of the relevant features associated with transitive events is that the agent
is not affected by the event in any salient way, but the effect manifests itself only in
the change-of-state of the patient.102 Consequently, we expect to find languages in
which the affectedness of agent results in a transitivity decrease. A couple of
examples are illustrated below, cf.

(1081) ga-nal qata bax-l-ej bu-r

he-ERG house.ABS 3CL.sell-DUR-CON.PRES 3CL.AUX-3SG
‘He has sold the house’
(1082) ga qata bax-l-ej u-r
he.ABS house.ABS 3CL.sell-DUR-CON.PRES 1CL.AUX-3SG
‘He has sold the house’ (Lak, Kazenin 1998: 112)

(1083) kati õa-tu tayiõa

meat.NOM I-ERG eat-FUT
‘I’m going to eat the/some meat’
(1084) kati õa-nyi tayiyindriõa
meat.NOM I-NOM eat.REFL.FUT
‘I’m going to have a feed of meat’(Yandruwandha, Breen 1976: 595)

102
It has to be stressed that in practice the agent is somehow affected by most transitive events.
Typical examples include, for example, ‘eating’ and ‘drinking’ both of which result in a condition
of not being hungry or thirsty. The more salient effect is, however, on the patient, which is important
for our purposes.
382

(1085) t~y kuzantaiyai anai-tt-~l

mother.NOM child.ACC embrace-STRONG PAST-AGR
‘The mother embraced the child’
(1086) t~y kuzantaiyai anain-t-~l
mother.NOM child.ACC embrace-W EAK PAST-AGR
‘The mother embraced the child’ (Tamil, Klaiman 1988: 43)

In the two first languages, the latter example is more appropriate, if the event has
had some effect on the agent. In Lak, the ergative construction only states that the
house has been sold, while the bi-absolutive construction exemplified in (1082)
implies that the selling made the agent homeless or rich, for example (Kazenin:
ibidem). In similar vein, (1084) adds the nuance that the agent is involved in the
event for his own benefit (Breen: ibidem). In Tamil, the relevant factor (in this
particular case) is whether the agent- or the patient-like participant is viewed as
more directly affected by the event.103 Example (1085) illustrates the typical case,
whereas in (1086) the event affects the agent more severely.
The agents in events profiled above acts volitionally and also controls the
event. Controlling agents that act less volitionally can be regarded as the second
mildest deviation from the agent prototype. Causees in causativized transitive events
are the most obvious examples of this (Lee (1985: 147) has labelled causees as
‘enforced agents’, but the label is applicable to cases below as well). The
introduction of a ‘higher agent’ deprives the agent of complete volitionality. A
couple of different examples are given below, cf.

(1087) mies raken-si talo-n

man.NOM build-3SG.PAST house-ACC
‘The man built the house’
(1088) mieh-en täyty-i rakentaa talo
man-GEN/ACC must-3SG.PAST build.INF house.NOM
‘The man was forced to build a house’ (Finnish)

(1089) a/egy férfi le-mos-t-a az autó-t

ART man PVB-wash-PAST-3SG ART car-ACC

103
The weak vs. strong past marker is also a marker of genuine transitivity, see (1214)-(1215).
 383

‘A/the man washed the car’

(1090) a/egy férfi-nak le kell mos-ni-a az autó-t
ART man-DAT PVB must wash-INF-3SG.POSS ART car-ACC
‘A/the man had to wash the car’ (Hungarian, examples courtesy of Magdolna Kovács)

(1091) aharen doru leppin

I door close.PAST.3SG
‘I closed the door’
(1092) ahannaš doru leppunu
I.DAT door close.INVOL.PAST
‘I closed the door (involuntarily)’ (Dhivehi, Cain & Gair 2000: 57)

(1093) sang.nyin ngas las.ka ‘di byed.kyi.yin

tomorrow I.ERG work this do.LINK.AUX(VOL)
‘I shall do this work tomorrow (of my own free will)’
(1094) sang.nyin nga las.ka ‘di byed.kyi.red
tomorrow I work this do.LINK.AUX(INVOL)
‘I shall be doing this work tomorrow (whether I like it or not)’ (Tibetan, Denwood 1999: 137f)

Examples illustrated above present cases in which events involving enforced agents
are marked differently from typical transitive ones. Changes in volitionality do not
affect the control parameter in any (significant) way, but the agent is completely
controlling the given events. In all the languages above, changes are expressed by
manipulating the marking of Agent.
Typical agents plan their actions in advance (see the division of transitive
events in four different phases in 3.2.1.). They are aware of the consequences of
their actions and know what they are going to do. Hence, events that are instigated
spontaneously (even if consciously) illustrate deviations from the transitive
prototype. The deviation is semantically rather insignificant, but there are languages
in which the spontaneity of actions produces a transitivity alternation, cf.

(1095)ravii-ne davaaii pii daalii

Ravi-ERG medicine.NOM drink pour.PERF
‘Ravi (deliberately) drank up the medicine’
(1096) ravii davaaii pii gayaa
Ravi.NOM medicine.NOM drink go.PERF
‘Ravi (impulsively) drank up the medicine’ (Hindi, Mohanan 1994: 74)
384

(1097) mam c ee wacane kiwwa

I.NOM that word say.PAST
‘I said that word’
(1098) mat c ee wac cne kiycwuna
I.DAT that word say.PASS.PAST
‘I blurted that word out’ (Sinhala, Gair 1990: 17)

In Hindi, the ergative marking of the Agent (in past tense and perfective aspect) is
related to high agency (of human agents). The event in (1095) can be considered
highly agentive. On the other hand, the agency is somewhat lower in (1096), which
is reflected in the omission of the ergative marking. Moreover, the verb morphology
is modified. The marking relations are converse in Sinhala, since the Agent is
marked explicitly in constructions denoting spontaneously instigated events. In
(1095)-(1098), the only difference between events denoted is the spontaneity of
(1096) and (1098). The actions are controlled by agents.
Very closely related to planning of events is that the agent is acting
consciously. Hence, in some languages typical transitive constructions can be used
only if the action of the agent can be regarded as completely conscious, cf.

(1099) kainzi £alO Jaq’-Jal-in (oqui-go)

grape.PL PVB eat-INTR-AOR (3SG-ALL/LOC)
‘He ate the grapes unconsciously’ (Batsbi, Holisky & Gagua 1994: 197)

(1100) ’ctt-cn ine-nlcp’et-g’i gutilg-e

dog-ABS ANTIP-break-3SG leash-INSTR
‘The dog broke its leash (and ran away)’
(1101) ’ctt’-e r clcp’en-nin gutlig-cn
dog-ERG break-3SGI.3SGII leash-ABS
‘Id.’ (Chukchi, Polinskaja & Nedjalkov 1987: 263)

As the translation of (1099) shows, the agent is acting unconsciously. Unfortunately,

the authors do not explicate the nature of unconsciousness more closely. A possible
interpretation might be that (1099) is appropriate in contexts in which the agent is
 385

lost in thoughts and is not concentrating on the eating.104 In this kind of case, the
action occurs more or less automatically. Examples (1100) and (1101) illustrate the
less transitive marking of agents that are inherently less aware of the consequences
of their actions. The event denoted by both examples is the same. According to
Polinskaja & Nedjalkov (ibidem), the first construction is the far more natural one,
since it does not imply that the agent would have acted consciously. The transitive
construction illustrated in (1101) is odd, since it implies that the dog is aware of the
consequences of its actions and acts consciously. A somewhat similar alternation is
found in Finnish (perhaps among other languages). In Finnish, the use of intransitive
vs. transitive reflexives is partially determined by whether the profiled event is seen
as occurring automatically or whether the agent is really making an effort to cause
the event to occur. In the former case, the intransitive reflexive is preferred. For
example, clauses like ‘I dressed’ refer to frequently occurring instances of dressing.
On the other hand, the use of a transitive reflexive (‘I dressed myself’) add the
nuance that the agent is really making an effort to causing the dressing event to
happen. This is more appropriate in case the agent has been sick for a very long time
and is still weak. Hence, that the agent can dress him/herself requires a real effort
and the agent is more involved in the event profiled.
Different aspects of intentionality and control are in general very important for
agency. Agents that lack either (or both) of these two features are clearly less typical
instigators of events than prototypical agents. Intentionality and control are rather
closely related and they are multilayered notions, which increases the number of
possible alternations due to changes in these two parameters. Various alternations
of this kind are illustrated and discussed in what follows.
The first type under study is presented by transitive events targeted at a
‘wrong’ (kind of) target. The agent completely controls the event, but the patient is
not the intended one, cf.

(1102) lam cya bat kææva

child.NOM rice eat.PAST.ACT
‘The child ate rice’

104
This analysis is entirely our own and it might be completely unjust to the data.
386

(1103) lam cya-šc vaha kævuna

child-DAT poison eat.PAST.PASS
‘The child (accidentally) ate something poisonous’ (Sinhala, W ijayawardhana et al 1995: 108)

(1104) saya minum kopi saudara

I drink coffee your
‘I drank your coffee’
(1105) maaf, kopi saudara ter-minum oleh saya
sorry, coffee your ACCID-drink by me
‘Sorry, I accidentally drank your coffee’ (Indonesian, De Vries 1983: 157f)

Second examples in both pairs denote an action targeted at a ‘wrong’ patient.

‘Eating’ and ‘drinking’ per se are controlled events (cf., however (1099)). Hence,
the nature of these events excludes fully accidental readings (cf. IAC’s below).
Consequently, the only meaningful interpretation of (1103) and (1105) is that the
target was not the right one. For example, what makes (1105) ‘accidental’ is that ‘I’
drank ‘your’ coffee instead of ‘mine’ (cf. also (402) from closely related Malay).
A further important facet of intentionality is that the result of the event
coincides with the intention of the agent.105 This means that the agent must be aware
of the consequences of his/her actions and s/he expects that the result of the event
corresponds to this. For example, the intentionality is very low in the case where an
agent who hits someone with no intention to hurt him/her, happens to kill the person
in question. The action by the agent is completely controlled, but the result is
radically different from that intended by the agent. An example in which this might
be the primary conditioning factor of the transitivity alternation is illustrated in
(1106), cf.

(1106) aharen(ge) at-un doru leppunu

I(GEN) hand-INSTR door close.INVOL.PAST
‘I closed the door (accidentally)’ (Dhivehi, Cain & Gair 2000: 57f)

Even if not explicitly stated by the authors, (1106) could denote an event in which
the agent keeps on swinging the door back and forth without intending to close it,

105
I thank Scott DeLancey for drawing my attention to this point.
 387

happens to do that. In this case, the result of the action disagrees with the intention
of the agent. Since genuine examples of this alternation have been notoriously
difficult to find, it seems that the given feature is understood as an integral part of
intentionality in general and is not usually viewed as an independent feature.
Intentionality is rather closely related to control. A complete lack of control
usually excludes intentionality as well (but not vice versa, as shown above). A
prototypical agent controls the event during its whole duration. Examples (1102)-
(1106) illustrate purposeful instigation of events. However, the agent does not fully
control the completion. A couple of further examples of this are illustrated below,
cf.

(1107) utang-ku ana sing durung ke-bayar

debt-my exist which not.yet pay-patient=subject-accidental
‘I haven’t had a chance to pay all my debts yet’ (Javanese, Poedjosoedramo 1986: 37)

(1108) cy-n c m chak-pu u-d-ri

I-ERG he-ACC see-NEG-NFUT
‘I did not see him’
(1109) cy-bu m chak u-d-ri
I-ACC he see-NEG-NFUT
‘I myself did not see him’ (Manipuri, Bhat & Ningomba 1997: 112)

(1110) a-psaasa-m q yaana-ha-r Ø-q’a-y-a-d c-ha

the-girl-OBL shirt-PL-ABS 3-HOR-3-PRES-sew-PL
‘The girl is sewing the shirts (completing them)’
(1111) q yaana-ha-m a-psaasa-r Ø-q’a-y-ha-a-w-d c:a-wa
shirt-PL-OBL the-girl-ABS 3-HOR-3-DAT-PROG-sew.INTR-PRED
‘The girl is busy sewing at the shirts (trying to sew them)’ (Kabardian, Colarusso 1992: 135)

Above, the agent is controlling the denoted events, but fails to carry it out due to
reasons beyond his/her control (cf. also (1039)-(1042) above). In Javanese, an
involitive (passive) affix is attached to the verb to underline this. In this particular
case, the person referred to has not been able to pay his/her debts, for which s/he is
not alone to blame for. In Manipuri, the transitivity decrease is more obvious, since
the marking of Agent shifts from ergative to accusative and that of Patient from
388

accusative to zero (see (911)-(913) from Central Pomo for a similar alternation).
According to Bhat & Ningomba (ibidem), the construction type illustrated in (1109)
is used, if the agent ‘had failed to carry our the expected activity due to reasons that
were not under his control’. (1110) denotes an event in which an agent is
successfully completing a transitive event. Example (1111), on the other hand,
describes a transitive event the agent cannot complete successfully. Also in this
case, the agent is controlling what she is doing, but her own inability causes the
event to be less fully completed than expected. The agent does not control the event
as a whole.
As noted, the agent fully controls his/her own actions in the events denoted
above. It, however, fails to carry out the desired activity. What is also important is
that the participation of the agent in events can be regarded as volitional and
purposeful. The lack of control can also be related to the initiation phase. This
usually results in a complete lack of control. Moreover, it implies that the
participation of the agent in an event cannot be seen as purposeful or volitional, but
is rather accidental and involuntary. Control usually implies intentionality, since we
can exercise control only if we know what we are doing. The deviation from the
transitive prototype is more obvious in the latter kind of case exemplified below, cf.

(1112) nyurra guni-ny ngayany

2SG.A cut-PAST self
‘You cut yourself’ (intentional action)
(1113) nganya guni-yi-ny
1SG.O cut-INTR-PAST
‘I cut myself accidentally’ (Djabugay, Patz 1991: 300)

(1114) ngatha ngunha-ngku yawarta-lu thala-nnguli-nha

1SGNOM that-AGT horse-AGT kick-PASS-PAST
‘I got kicked by a horse’
(1115) ngatha ngunha-mari yawarta-mari thala-nnguli-lha
1SGNOM that-CAUSAL horse-CAUSAL kick-PASS-PERF
‘I got a kick from the horse’ (Panyjima, Dench 1991: 195)

(1116) bhurus jhuwa n-gekeq hasan

dog that AV-bite Hasan
 389

‘The dog bit Hasan’106

(1117) dlubang jhuwa ka-obbhar (biq) hasan
paper that IN-burn (by) Hasan
‘Hasan accidentally burned the paper’ (Madurese, Davies 1999: 29)

(1118) õa-s dkaryol bcaq-pa-yin

I-ERG cup break-PERF-VOL
‘I broke the cup purposefully’
(1119) õa-s dkaryol bcaq-so õ
I-ERG cup break-PERF.INVOL
‘I broke the cup accidentally’ (Lhasa Tibetan, DeLancey 1984: 132)

(1120) na-ta-ngieng ureueng-nyan lê-gata baroe

BE-2-see person-that by-you yesterday
‘You saw that person yesterday’
(1121) na-teu-ngieng'-teuh ureueng-nyan lê-gata baroe
BE-DC-see-2 person-that by-you yesterday
‘You accidentally saw that person yesterday’ (Acehnese, Durie 1985: 60f)

Examples above illustrate ‘involuntary agent constructions’ (henceforth, IAC for

short) as defined by Haspelmath (1993:291). As the label implies, the agent is
involuntarily (or accidentally) partaking in an event. It is an agent without intending
to be one, which naturally reduces the agency. The agent might be involved in
another activity completely controlled by it, but the control (or volitionality) is not
extended to the event denoted by an IAC. For example, in (1112) the agent is
volitionally carrying out a controlled activity that causes itself to be cut. In (1113),
on the other hand, the agent is not a volitionally acting or controlling agent in the
same event. The agent may be, for example, carrying an axe around that at some
point cuts him/her. The agent is fully controlling the carrying event, but the control
is not extended to the event of cutting. Other examples above are very similar as
regards the notions of control and volitionality. They are presented simply in order
to illustrate the structural variety of IAC’s.107 Examples from Acehnese are

106
This construction is used here simply to illustrate the basic transitive structure of Madurese, it is
not the agentive version of (1117).

107
For a more detailed typology, see Kittilä: submitted.
390

somewhat different. The differences between (1120) and (1121) are best interpreted
as being related to the planning of the event. (1120) is more appropriate, if the agent
(defined loosely) has agreed to meet with the person in question. S/he can control
the occurrence of the event per se. The use of (1121), on the other hand, implies that
the agent meets the person without having planned this. The event denoted cannot
be regarded as impulsive in the sense of (1096), since the occurrence as a whole is
accidental in (1121).
Very closely related to control is the notion of responsibility. The agent of
typical transitive events is construed as being directly (and alone) responsible for the
instigation of events. Constructions illustrated in (1087)-(1094) gave some foretaste
of alternations due to a lower degree of responsibility. The instigators of events
cannot be regarded as being completely responsible for the events in question, since
they are forced to act. Some further examples are given below, cf.

(1122) ha-yulang si juan i kareta

ERG.3SG-break UNM John the car
‘John wrecked the car’
(1123) man-yulang si juan gi kareta
ANTIP-break UNM John LOC car
‘John was involved/took part in wrecking of the car’ (Chamorro, Cooreman 1988: 580)

(1124) siriseen c gun cpaalctc hariy cta gæhuwa

Sirisena.NOM Gunapala.DAT really hit-PAST
‘Sirisena really beat Gunapala’
(1125) poliisiye õ gun cpaalctc hariy ctc gæhuwa
police.INSTR Gunapala-DAT really hit-PAST
‘The police really beat Gunapala’ (Sinhala, Gair 1990: 14)

(1126) di besun bitalo

I.A.ERG knfie.F.NOM lost.CAUS
‘I lost the knife (my fault)’
(1127) di ’o besun bito
I.C.OBL knife.F/NOM lost
‘I lost the knife (not my fault)’(Tindi, Kibrik 1985: 307)
 391

(1128) |íí dzaanééz yi-zta |

horse mule it.it-kick
‘The horse kicked the mule’ (the horse being responsible for what happened)
(1129) dzaanééz |íí bi-zta |
mule horse it.it-kick
‘The horse kicked the mule’ (the mule being responsible for what happened) (Navajo, Dixon
& Aikhenvald 1997: 98, cited from Hale 1973b)

Examples from Chamorro and Sinhala illustrate instances of ‘shared responsibility’.

Example (1122) implies that the agent is alone responsible for the event denoted.
The antipassivization has the function of reducing the degree of responsibility
associated with the agent. (1123) does not ‘free’ the agent completely from
responsibility, but it implies that the agent has not acted alone and is thus not alone
responsible for the event. Examples (1124) and (1125) are similar to those from
Chamorro. In (1124), Sirisena is construed as being directly (and alone) responsible
for the beating of Gunapala. Similarly to (1123), the degree of responsibility related
to an individual agent is somewhat lower in (1125). Several police officers might
have been involved in the beating or the order to carry out the action may come from
above. The position may force an individual to carry out the action in question.108
Examples from Tindi are very similar to IAC’s illustrated above. (1127) could
probably also be translated as he lost the knife accidentally. Examples from Tindi
underline the intimate relation of IAC’s and events involving agents not alone
responsible for events. Examples from Navajo are somewhat different. In (1128),
the agent is the responsible participant. In (1129), on the other hand, illustrates the
agent and the patient share the responsibility. It is more appropriate, if, for example,
the mule has done something to provoke the horse. The agent cannot be held entirely
responsible. This kind of alternation shares common features with the likes of
(1043)-(1048).
The use of typical transitive constructions usually implies responsibility from
the agent. Thus, in some languages the use of transitive constructions can be taken
as an insult or an accusation, which has resulted in the use of other coding strategies

108
Subjects referring to institutions appear in general in the instrumental case in Sinhala (see e.g. Gair
1998: 66ff).
392

instead (cf. e.g. Duranti & Ochs 1990: 17f). Samoan and Tongan are among the
languages that disfavour ergative Agents for this reason (see Duranti and Ochs 1990,
see also Knight 2001 for Bunuba). In ergative languages, the avoidance of ergative
Agents is easily understood, since the Agent is marked explicitly as bearing the role
of (responsible) agent (cf. also (1068)-(1080) above). The clause structure is,
however, manipulated for the same reason in other languages as well. Evidence for
this is provided by the frequent occurrence of IAC’s in the languages of the world.
As noted in 3.1.1., agency is in general a rather conceptual feature that cannot be
measured fully objectively. The number of features usually correlates with agency,
but each of these features is conceptual. We all conceptualize agency somewhat
differently. The rationale behind IAC’s is that the instigator of an event is not
purposefully initiating it and cannot be held fully responsible for it (even though it
has intitiated the event). Hence, we can employ IAC’s or fully intransitive
constructions, even if the agent is fully responsible for the event in order to avoid
an accusing nuance. Each of us must be familiar with the use of constructions like
it simply broke in contexts in which we wish to de-emphasize our responsibility.
The use of less transitive constructions does not have to correspond to an
ontological decrease in agency, but we only take advantage of the semantics related
to the construction.
Above we have illustrated alternations that follow inherently ((1052)-(1080))
or ‘situationally’ determined ((1087)-(1129)) lower degrees of agency. Languages
vary according to whether these are encoded similarly or not, cf.

(1130) õundu:n waguda õgu gunda:dinu

that.ERG man.ERG cut.di.PAST
‘The man cut (the woman) accidentally’

(1131) õanan ginga:õ giba:dinu

I.ACC prickle-ERG scratch.di.PAST
‘A prickle scratched me’ (Yidiñ, Dixon 1977: 275)

(1132) na tapuni e le matagi le faitoto’a

PAST close ERG ART wind ART door
‘The wind closed the door’
 393

(1133) na tapuni i le matagi le faitoto’a

PAST close LOC ART wind ART door
‘The wind closed the door’
(1134) na va’ai le fafine i le pule= ~’oga i le maketi
PAST see ART woman LD ART rule=school LD ART market
‘The woman saw the principal at the market (accidentally)’ (Samoan, Mosel & Hovdhaugen
1992: 423ff)

Yidiñ exemplifies a language in which these two semantically distinct alternations

are marked identically, whereas in Samoan they are distinguished
morphosyntactically. What makes the distinction between these two language types
relevant for our purposes is that the rationale behind these motivations is radically
different. In languages like Yidiñ, the use of typical transitive vs. less transitive
constructions is (among other things) conditioned by the less agentive nature of the
instigator in general. The precise motivation is irrelevant. It suffices that we are not
dealing with a typical agent. In Samoan, on the other hand, the motivation of lower
agency is relevant. In Samoan, both typical and less typical instigators of events can
appear in the basic transitive frame, but the different semantic changes are expressed
by distinct constructions. The specific semantics of (1133) and (1134) excludes the
other reading.
One final feature related to agency is illustrated by the directness of the
causation. Typical agents consciously target their actions at a patient, which directly
results in a change-of-state in the patient participant. The agent is also directly
responsible for the event in question. If the causation is less direct, the agency is
usually (but not always) adversely affected. We do not pursue this topic any further
in the present context, since causation (however indirect it may be) is construed as
a transitivizing alternation and will therefore be discussed in more detail below. It
is, however, important to note the effects of less direct causation on transitivity also
here, since there are cases in which it is structurally justified to claim that we are
dealing with a transitivity decreasing alternation. For the mere sake of simplicity,
causation as a whole will be dwelled on in 5.3.1.2.1.
Above, we have illustrated different kinds of transitivity alternation due to
decreased agency. It is also noteworthy in the present context that the nature of
events makes a significant contribution in this respect. In general, we may say that
394

events intrinsically related to high agency do not allow alternations that completely
de-emphasize the agency, cf.

(1135) zamara.di-waj get’e xa-na

Zamira-ADEL pot break-AOR
‘Zamira broke the pot accidentally/involuntarily’
(1136) *taibat.a-waj rak aq haj-na
taibat-ADEL door open-AOR
(Taibat accidentally opened the door) (Lezgian, Haspelmath 1993: 292)

(1137) er öffne-te das fenster

he.NOM open-3SG.PAST ART window
‘He opened the window’
(1138) *ihm ist das fenster geöffnet
he.DAT be.3SG.PRES ART window open.PARTIC
(He accidentally opened the window) (German)

(1139) ku-sala-pa-‘ita-‘e te boku

1SG-accident-CAUS-see-3OBJ CORE book
‘I accidentally showed them the book’
(1140) *ku-sala-hoko-leama-‘e
1SG-accident-FACT-good-3OBJ
(I accidentally really improved it) (Tukang Besi, Donohue 1999: 213)

In all the examples presented above, the latter constructions are odd at best (cf. also
(414)-(422) from Finnish and Malay). Hence, the use of IAC’s is restricted to certain
verb classes only (see Haspelmath: ibidem for a more detailed analysis of Lezgian).
Both in Lezgian and German, the verb ‘open’ is infelicitous in an IAC. This is due
to the inherently high degree of agency associated with this event. In similar vein,
(1140) from Tukang Besi is not acceptable, since it is rather difficult to imagine a
scene in which someone improves something completely accidentally.
Examples (1023)-(1134) illustrate cases in which either of the parameters
associated with high transitivity is primarily responsible for the alternations. The
other parameter has been more or less unaffected. Reciprocals exemplify kinds of
combination of certain aspects discussed above. As regards the number of
participants, reciprocals are always transitive constructions. They necessarily
 395

involve (at least) two participants. One manifestation of this is that a reciprocal
reading is possible only with plural Agents (cf. e.g. (533)). As opposed to transitive
events, both participants of a reciprocal situation are carrying out the same activity
(cf. e.g. Kemmer 1993: 96f (partly cited from Lichtenberk 1985: 21)). As a
consequence, similarly to reflexives, the agent itself is also affected by the event it
is involved in (which does not, however, affect the degree of agency in any way).
Furthermore, since reciprocals are obligatorily transitive events and the participants
are carrying out the same activity, the patient is not merely a passive participant. It
is important to note that the participants in a reciprocal scene must not be targeting
the action at an external participant, but both must also be affected by the event they
are involved in. The fact that both participants are simultaneously agents and
patients makes it rather difficult to definitely construe either of them as the agent
and the other as the patient. In typical transitive events, the inventory or roles is
always clear and we can consequently say which participant instigates the event and
which is affected by it. In a way, reciprocal events can even be thought of as two
distinct, yet very closely related transitive events. For example, the event ‘they hit
each other’ can be regarded as comprising two sequential instances of hitting (i.e.
‘A hits B’ and ‘B hits A’). This is, however, not necessarily the case, but the
participants may be active simultaneously and the affectedness may be gradual. This
is the appropriate reading, for example, in ‘they beat each other’. This event
involves two participants acting on each other simultaneously. Eventually, both get
affected by it. It is rather difficult (and also pointless) to distinguish between the two
events. Hence, the sequentiality of the two subevents is the less probable
interpretation in this case. This kind of reciprocal is best construed as a single
transitive event that, differently from typical transitive events, involves two
participants both of which can be viewed as active and affected. In the case of
sequential interpretation, the temporal distance between the two events is usually
(but not necessarily) relatively small. This does not have to be the case, however,
since two instances of ‘hitting’ separated by two weeks can also be construed as a
reciprocal event, even if the reciprocal becomes less probable and natural, if the two
events are separated by a very long interval. This speaks for the lower general
transitivity of reciprocals. In typical transitive events, the action by the agent and the
affectedness of the patient are integrated into one event. A couple of examples of
396

reciprocals are provided below109, cf.

(1141) tráak t vk §ò §
buffalo butt I
‘The buffalo butted me’
(1142) tráak tr-tvk y ]§ tèe
buffalo RECIP-butt each-other
‘The buffalos butted each other’ (Kammu, Svantesson 1983: 112)

(1143) scww-o -u irsb crs-a ew tc-d cbadd cb-u

person-PL-DEF each.other-POSS.3PL RECIP-hit.RECIP.PERF-3PL
‘The people hit each other’ (Amharic, Amberber 2000: 327)

(1144) pula pama kuce-ng minha pangku-la yenta

3DUNOM man two-ERG ANIMAL wallaby-3DUNOM spear
‘Those two men speared a wallaby’
(1145) pula pama kuce pulanmala-la yent-o
3DUNOM man two 3DUREFL-3DUNOM spear-RECIP
‘Those two men speared each other’ (Kugu Nganhcara, Smith & Johnson 2000: 426)

(1146) wa õgarany-dyu bama-õgu yanmira-ra-õa-dhana

all-ERG man-ERG laugh-RECIP-PRES-3PL.S/A
‘All men are laughing at each other’ (Biri, Terrill 1998: 43)

(1147) e loma-ni koya

he love-TRANS she
‘He loves her’
(1148) erau vei-loma-ni
they:two RECIP-love-TRANS
‘They two love each other’ (Fijian, Lynch 1998: 145)

Examples from Kammu illustrate a transitive reciprocal. Reciprocality is expressed

through a morphological change in the verb, in addition to which the Patient is a
reciprocal pronoun. In Amharic, reciprocal is less transitive than a typical transitive
construction, since no accusative affix can be attached to the reciprocal pronoun

109
Only morphosyntactic reciprocals concern us here, lexically expressed naturally reciprocal events,
like ‘fight’ (cf. Kemmer 1993: 102:ff) are excluded from the discussion.
 397

(Amberber:ibidem). The examples from Kugu Nganhcara are the mirror image of
Amharic, which results from the predominantly ergative nature of the language. The
Agent of a typical transitive construction appears in the ergative, while this is not
possible in reciprocals. Examples from Biri and Fijian illustrate cases in which the
number of arguments is affected. The intransitivization is incomplete in both
languages. In Biri, the Agent appears in the ergative also in reciprocals, while in
Fijian the transitive affix is retained. The less complete intransitivization is rather
natural, since reciprocals profile events construed as transitive on the basis of the
number of participants involved (cf. also Rumsey 2000: 119 for Bunuba).
A further interesting aspect associated with reciprocals is illustrated below, cf.

(1149) o sepo e’ oti-va ti’o a ulu-i Elia

Sepo cut-TRANS CONT
‘Sepo is cutting Elia’s head (hair)’
(1150) erau sa vei-’oti ti’o sepo vata ’ei elia
3DU ASP PR-cut CONT Sepo together with Elia
‘Sepo and Elia are involved in an activity of (hair) cutting’ (Boumaa Fijian, Dixon 1988: 177,
the analysis of the latter example from Lichtenberk 2000: 37)

Examples above emphasize the importance of an obvious semantic role inventory

for transitivity. Only cases in which we can unequivocally distinguish between agent
and patient are considered highly transitive in Fijian. Above, this is the case only in
(1149). According to Dixon (ibidem) the latter construction is ambiguous as regards
the semantic role assignment. In terms of Kemmer (1993: 97), the degree of event
elaboration is much lower in (1150). The roles can, however, be disambiguated by
the context. These cases can be deemed as a special case of reciprocal, since the
event itself is not necessarily reciprocal despite being encoded as one.

5.3.1.1.2. De-transitivizing alternations

In the previous sections, we have examined alternations that are motivated by

obvious semantic changes. We simply need to refer to ontological differences
between non-linguistic events to explain the occurrence of these alternations. It is
of the utmost importance to distinguish between instances of basically the same
398

events. Most intransitivizing alternations exemplified above serve this function. If

the differences are not contextually inferable, we have to resort to linguistic cues.
In this section, we proceed to alternations that cannot be explained by the need to
distinguish different instances of the same events from each other. The alternations
are not due to ontological differences and are not non-linguistically salient. The
resulting less transitive constructions are not non-linguistically distinguishable (for
example, he eats the meat and the meat was eaten by him can both refer to the same
event, whereas he washed the child and the child washed (himself) cannot). As a
result, these alternation types should be kept apart despite the obvious structural
overlaps (see below). Hence, the term ‘de-transitivization’, the use of which is
restricted to cases in which ontological and structural transitivity disagree and in
which the ontological transitivity outranks the structural one. Hence, as de-transitive
are classified derived intransitives whose use is not conditioned by differences
between distinct events. Typical de-transitive alternations result in a change in the
number or status of arguments, but these do not correspond to changes in the
number or semantics of participants. De-transitive alternations are made possible by
the semantics of verbs. This means that, since the number (and nature) of
participants can be inferred from the semantics of a verb, the participants do not
need to be explicitly referred to. De-transitive alternations can also be divided into
two on the basis of whether the alternations affect the number of arguments or not.
However, because of the ‘non-ontological’ motivation, this distinction does not
coincide with the one above, i.e. the number of participants involved remains the
same in all cases. We begin by illustrating the more obvious de-transitivizing
alternations in that the number of arguments is affected.
As noted above, de-transitivizing alternations differ fromthe intransitivizing
ones insofar as we cannot justify their inclusion in the typology by referring to
obvious changes in semantic transitivity. However, what entitles us to take account
of de-transitivizing alternations is that de-transitive alternations structurally
resemble many less than perfectly transitive constructions. This is very obvious, if
de-transitivization affects the number of arguments. (cf. e.g. he eats and he runs).
We rather proceed from structure in these cases. De-transitivizing alternations can
be taken account of, since the resulting structure is so obviously less transitive than
typical transitive ones. On the basis of structural criteria (e.g. morphological
 399

marking of the verb), inverse constructions are excluded despite their obvious
functional similarity to passives (cf. e.g. Thompson 1994: 61f, Jacobs 1994: 127).
The latter also applies to languages in which highly transitive events can be
expressed in more than one different way without any semantic motivation (a
possible example of such a language is provided by Toratán, cf. Wolff and
Himmelmann 1999).
As illustrated in the previous section, the number of semantically different
intransitivizing alternations is rather high. First of all, the distinction between
alternations affecting the number of participants and those that only affect some
individual transitivity features is relevant. Second, the number of individual features
that can result in a transitivity alternation is high. For example, as illustrated above,
a number of distinct agency features can cause a transitivity alternation. Both of
these aspects are irrelevant for the notion of de-transitivization. This inheres in the
definition outlined above. Since de-transitivization must not impinge on the nature
of the basic event itself, all of the criteria that distinguish between different
intransitivizing alternations are by definition irrelevant here. Because the number
of conditioning factors of de-transitivizing alternations is considerably lower, the
focus of the presentation below lies on the structure. We are also concerned with the
rationale behind the structural variety. As opposed to a semantic (or ontological)
motivation, typical de-transitive alternations are conditioned by significant changes
in the definiteness (understood in a very broad sense) of participants referred to. In
a great number languages, only participants relevant in a given situation are
explicitly referred to or are encoded by a typical transitive construction. Participants
whose identity is not important can be omitted rather freely. Consequently, passives
and antipassives present the most obvious examples of de-transitivizing alternations.
Typical examples are illustrated below, cf.

(1151) zall§an a-t-i kawo-z-aa wod’-i-d-a

merchant-PL-SUBJ king-D-OBJ kill-PM-TENSE-X
‘Merchants killed the king’
(1152) kawo-z-ii zall§an a-t-a-n wod’-ett-i-d-es
king-D-SUBJ merchant-PL-OBJ-LOC kill-PASS-PM-TENSE-X
‘The king was killed by merchants’ (Gamo, Éva 1990: 394)
400

(1153) danny-ngka kuldji dara-y-mani-n

Danny-ERG stone fall-NONCMPL-TRANS-CMPL
‘Danny dropped the stone’
(1154) danny-nda kuldji dara-n
Danny-LOC stone fall-CMPL
‘The stone was dropped by Danny’ (Guugu Yalandji, Blake 1977: 26)

(1155) ar-en ×um-ax aq’-i-ne

son-ERG bread-DAT take-AOR-3SG
‘The son took the bread’
(1156) ×um-Ø aq’-ec-i-ne
bread-ABS take-AUX(INTR)-AOR-3SG
‘The bread was taken (*by someone)’ (Udi, Schulze-Fürhoff 1994: 497)

(1157) ta-hoqi rao

PASS-gore 1SG
‘I was gored’
(1158) ta-hoqi-ni-a rao sa boko
PASS-gore-APPL-3SG 1SG ART.SG pig
‘I was gored by that pig’
(1159) ta-nani leboto sa malego tani sa
PASS-bite machete ART.SG leg 3SG.POSS 3SG
‘Her leg was cut by a machete’ (Hoava, Davis 1997: 261)

(1160) tuttu taku-aa

caribou see-3SG.3SG.IND
‘He saw the caribou’
(1161) tuttu-mik taku-vuq / taku-nnip-puq
caribou-INSTR see-3SG.IND / see-ANTIP-3SG.IND
‘He saw a caribou’ (W est Greenlandic, Fortescue 1984: 86)

(1162) wane e laba-tani-a maka nau

man 3SG.NFUT harm-TRANS-3SG.OBJ father 1SG
‘The man harmed my father’
(1163) roo wane kero kwai-laba-taqi
two man 3DU:NFUT DEPAT-harm/spoil/etc-TRANS
‘The two men harm (people), spoil, damage (things) etc.’ (Toqabaqita, examples courtesy of
Frank Lichtenberk)
 401

Examples (1151)-(1159) illustrate different kinds of passive, whereas (1160)-(1163)

exemplify antipassives. The illustrated examples differ from each other in the nature
of de-transitivization. In Gamo, the passivization is expressed by manipulating the
verb morphology, which is accompanied by changes in the status (and possibly
number) of arguments. Guugu Yalandji lacks a specific passive morpheme, but
passivization is signalled by omitting the transitive verb morphology. Udi differs
from Gamo and Guugu Yalandji, since the passive is obligatorily Agentless. In
Hoava, passivization excludes the expression of Agent, but a non-human Agent can
be expressed, if the verb is applicativized. As (1159) shows, this applies only to
definite Agents, since indefinite agents can be referred to without applicativization.
Examples from West Greenlandic illustrate morphological antipassives that affect
the status of arguments. The morphological de-transitivization (i.e. attaching the
affix -nnip- to the verb) is optional, as shown in (1161). Examples from Toqabaqita
illustrate the mirror image of (1155) and (1156), since de-transitivization makes it
impossible to express a Patient. Cases like (1156) and (1163) are structurally similar
to intransitivizing alternations that decrease the number of participants. The
difference between these two kinds of alternations is that only in intransitivizing
alternations the number of arguments is semantically determined. In (1156) and
(1163), on the other hand, the structure makes it impossible to refer to agent or
patient explicitly.
Constructions presented in (1151)-(1163) illustrate what is typically regarded
as passive (see e.g. Siewierska 1984: 2, Shibatani 1985: 837 and Keenan 1985: 247,
for a different approach see Kittilä 2000) or antipassive (e.g. Dixon & Aikhenvald
1997: 73). In these cases, the possible decrease in valency is accompanied by a
morphological manipulation of the verb. The argument omission is, however,
possible also without any kind of explicit de-transitivization of the predicate, as has
already been discussed above. These are usually excluded from studies of passives
and antipassives, since they lack one of the most important criteria, i.e. the verb
morphology is not de-transitive. Since the basic function of the two cases (i.e. the
omission of reference to ‘irrelevant’ participants) is the same, the kinds of
alternation illustrated below are taken account of here, cf.
402

(1164) balan dugumbil (ba õgul ya a-õgul) balgan

ART.ABS woman.ABS (ART.ERG man-ERG) hit
‘Man (someone) hits woman’ (Dyirbal, Van Valin 1977: 691, cited from Dixon 1972)

(1165) er ha-t (fleisch) gegessen

he.NOM have.PRES-3SG (meat(ACC)) eat.PARTIC
‘He has eaten (meat)’ (German)

(1166) ada (jak) t’üna

he.ERG (meat.ABS) eat.AOR
‘He ate the meat’
(1167) jak t’üna
meat.ABS eat.AOR
‘The meat was eaten’ (Lezgian, examples courtesy of Martin Haspelmath)

(1168) na’e inu ‘a e kavá ‘e Sione

PAST drink ABS the kava ERG John
‘John drank the kava’
(1169) na’e inu (kavá) ‘a Sione
PRET drink (kava) ABS John
‘John drank (kava)’
(1170) na’e inu ‘a e kavá
PAST drink ABS the kava
‘The kava was drunk’ (Tongan, Churchward 1953: 76) 110

Example (1165) illustrates the passive analog in an ergative language derived from
the corresponding transitive construction simply by eliminating the Agent. There are
no morphological changes in the verb and the marking of the remaining argument
is not affected. The German example illustrates the mirror image of this, since the
function of antipassive is expressed simply by omitting the Patient. Lezgian and
Tongan exemplify languages in which both passive and antipassive functions are
expressed by eliminating either argument without modifying the verb morphology.
Lezgian is a kind of combination of Dyirbal and German, since both arguments of
transitive constructions are freely omissible. What makes Tongan somewhat
different in this respect is that the Patient omission results in the promotion of the

110
The last example has been constructed by us based on data from elsewhere.
 403

Agent in status. Differently from typical antipassives, however, the verb morphology
remains constant.
Passives and antipassives are de-transitivizing alternations that typically affect
the number of arguments. In most languages, passives and antipassives are
optionally valency decreasing alternations. Another de-transitive alternation that
reduces the number of independent arguments is provided by incorporation. As
discussed in 5.2.3., incorporation differs from patient deleting antipassives in that
the patient is explicitly referred to despite having lost its status as an independent
argument. What these two constructions (i.e. antipassives and noun incorporations)
do have in common is that they are both conditioned by the irrelevance of
participants referred to. As opposed to antipassives (defined in a very broad sense
including also cases like he eats), incorporation obligatorily identifies the patient,
irrespective of how indefinite it is. A couple of examples of different noun
incorporations are illustrated below, cf.

(1171) a-k cm a-ho-j

he-make he-hole-DO
‘He digs his hole’
(1172) a-ho-k cm
he-hole-make
‘He hole-digs’ (Kitonemuk, Mardirussian 1975: 383)

(1173) g cm-nan m cng-uwwi t-ilg ctav-Ø-cna

1SG-ERG hand.ABS.PL 1SG.A-wash-AOR-3PL.P
‘I washed my hands’
(1174) g cmm c tc-m cng-ilg ctav-Ø-ck
1SG.ABS 1SG.S-hand-wash-AOR-1SG.S
‘I washed my hands’ (Alutor, Koptjevskaja-Tamm & Muravyova 1993: 298)

(1175) manuhor jabu i ahu

buy house DET I
‘I buy the house’
(1176) hu-tohor jabu i
1SG-buy house DET
‘The house is bought by me’ (Toba Batak, Keenan 1985: 264, examples from S. Mordechay)
404

Examples above illustrate three somewhat different kinds of incorporation. In

Kitonemuk, the Patient is incorporated into the verb, which results in the omission
of the direct object marker. The incorporation has more drastic consequences for the
clause structure in Alutor. The explicit ergative marking on Agent is omitted, in
addition to which the verb agreement is intransitivized. Toba Batak differs from
Kitonemuk and Alutor in that the Agent, and not the Patient, is incorporated into the
verb. Both in Kitonemuk and Toba Batak, the verb morphology remains constant.
Incorporation differs from passives and antipassives in their status as genuine
de-transitivizing alternation. Both passives and antipassives have also in many
languages functions not related to typical de-transitivization. For example, in Yidiñ
and Diyari (see (512)-(517)), a general passive or antipassive construction also
expresses many semantic functions (such as reduced agency). Incorporation, on the
other hand, seems to be more intimately related to de-transitivization only.111 The
primary function of incorporation is to reduce the status of arguments. We have not
been able to find examples of cases in which incorporation would be obviously
related to the expression of changes in ontological transitivity (e.g. languages in
which less affected patients are obligatorily incorporated into the verb). What is of
further interest is that incorporation usually reduces the status of Patients, which
makes them functionally more similar to antipassives than passives (languages like
Toba Batak are cross-linguistically rare). Of the two de-transitivizing constructions
illustrated above, antipassive is the one that also has intransitivizing functions.
Antipassive is also related, for example, to the expression of imperfective aspect or
lower degree of affectedness in many languages. Since incorporation does not seem
to have these kinds of function, we may conclude that we are dealing with two
functionally overlapping, yet separate constructions. Incorporation is functionally
more restricted, which makes it the functionally purest de-transitivizing alternation
type.
Alternations illustrated above radically affect the status of arguments.
Moreover, there are also alternations that do not eliminate arguments, but only
decrease their status. The resulting constructions are typically syntactically
transitive, but morphologically intransitive. A couple of somewhat different

111
This is a preliminary claim that is not based on a detailed study of any kind.
 405

examples are illustrated and discussed below, cf.

(1177) ketâb-ra xând-am

book-POSTP read/PAST-1SG
‘I read the book’
(1178) ketâb xând-am
book read/PAST-1SG
‘I read a/some books’ (Persian, Lazard 1998: 168)

(1179) ama ondu manusa-na nooDida

he a man-ACC see.PAST.3SG
‘He saw a man’
(1180) ama ondu kaTTE baNDi(-ya) nooDida
he a wood vehicle(-ACC) see.PAST.3SG
‘He saw a waggon’ (Badaga, Lazard 1998: 189, cited from Pilot-Raichoor 1991, 1994)

(1181) krar-w kwr py y-wr-y

dog-ERG pig that.ABS OBJ-bite-PAST.SG.SUBJ
‘The dog bit the pig’
(1182) krar kwr (*py) y-wr-y
dog pig (that.ABS) OBJ-bite-PAST.SG.SUBJ
‘The dog bit pigs’ (Kanum, Donohue 1997: 8f)

(1183) e fena §a-ni / (* ) fanu Ø-yau-a-§a

DEM woman-ERG / (*ABS) man 3SG-see-3SG-IND
‘That woman saw the man’
(1184) e fena §a-ni /() pai yau-a-§a
DEM woman-ERG / (ABS) pig see-3SG-IND
‘That woman saw the pig’ (Tauya, MacDonald 1990: 120f)

(1185) ko-i tuhi-a te puke ko ino

PERF-3SG cut-TRANS ART puke PERF fall
‘(He) cut down the puke-tree, and it fell’
(1186) io-ko mhe-la ko tãtuhi talie ala
CONJ-TOP boy-DEM PERF REDUP.pick talie PART
‘And the boy went on picking talienuts’ (Pileni, Naess 2000: 51)

Examples (1177)-(1180) illustrate very typical examples of ‘differential object

406

marking’ (term adopted from Lazard 1998:82). In Persian and Badaga, definite
Patients are marked differently from indefinite ones. Only definite Patients appear
in the accusative, as in (1177) and (1179). The two languages diverge in that in
Persian the postposition râ is omitted altogether in the case of indefinite Patients,
whereas in Badaga the accusative becomes optional. Kanum and Tauya illustrate a
kind of mirror image of Persian and Badaga. The ergative marking of Agent is the
main indicator of Patient definiteness. The manifestation is different from (1177)-
(1180), which results from the basic argument marking pattern. Kanum is more
similar to Persian, since the omission of ergative marking of Agent implies that the
Patient is indefinite. As shown by Donohue, ergativity is clearly related to Patient
definiteness, since in cases like (1182) it is not possible to modify the Patient by
determiners. In Tauya, this is not necessarily the case, since the Agent in clauses like
(1184) can also appear in the ergative. However, differently from clauses involving
definite Patients, the ergative case marking is optional. Consequently, Tauya is more
similar to Badaga. Languages like Kanum and Tauya underline the relevance of
Patient definiteness for high transitivity in general. In these two languages, the use
of an ergative construction requires that the Patient be definite. That there are
languages like Persian, Badaga and also Turkish, Spanish etc. is more expected,
since the changes in definiteness are expressed iconically (cf. (712)-(723)). Pileni
illustrates a de-transitivizing alternation in which the verb morphology is de-
transitivized without this resulting in any changes in valency. The Patient has to be
indefinite, which in this case means that it is not preceded by a definite article (for
a somewhat different reflection of indefiniteness, see Lichtenberk 1982: 264).
Above, we exemplified constructions in which de-transitivization causes
obvious changes in the argument marking. Differently from passives and
antipassives, the number of arguments remains constant. The last type of de-
transitivizing alternation discussed here is presented by cases in which the status of
arguments is demoted without this affecting the number or marking of arguments
in any significant way. These include constructions in which the Agent or the
Patient is referentially demoted and is referred to by a generic particle or pronoun,
cf.
 407

(1187) er / man trink-t bier da

he.NOM / ‘they’NOM drink.PRES-3SG beer there
‘He drinks/they drink beer there’
(1188) he / ?man ass den apfel
he.NOM /‘ they’.NOM eat.PAST.3SG ART.ACC apple
‘He/?they ate the apple’ (German)

(1189) tawa llëbo yo-no

1PL.EXCL things eat-PERF
‘W e ate things’
(1190) tawa yo-no llë
1PL.EXCL eat-PERF MASS
‘W e ate’ (Teribe, Quesada 2000: 145)

(1191) sake a-ku

sake 1SG-drink
‘I drink sake’
(1192) i-ku-an
Generalized object-drink-1SG
‘I drink’ (Ainu, Shibatani 1990: 46)

Examples (1187)-(1188) illustrate the man-construction of German (attested in a

very similar form also in Swedish). Functionally, this construction is close to the
typical passiv, since its main function is to demote the Agent. Differently from er,
man does not refer to an identifiable agent. What makes this construction different
from passives is the structural transitivity. The verb is not de-transitivized nor is the
morphological marking of arguments affected. A further interesting point is
illustrated in (1188). The primary function of man-construction is in German to
demote the Agent (only referentially). However, the Patient is also usually implicitly
demoted, since Patients of man-constructions are typically (yet not obligatorily)
indefinite. Examples from Teribe and Ainu illustrate the mirror image of (1187)-
(1188). The non-referential argument is the Patient. The translation given by
Quesada underlines the de-transitivizing effects of the given marker. Identically to
German, the marking of the verb is not affected. In Ainu, the genericity of the
Patient is signalled by attaching a generalized object affix to the verb. The
mechanism is rather similar to antipassivization.
408

Above, we have illustrated four different kinds of de-transitivizing alternation.

The alert reader may object to the rather detailed structural typology proposed, since
at the beginning of the section we stressed that the structural nature of functionally
identical alternations is irrelevant. The illustration above contradicts this. The
rationale behind the structural orientation of the section lies in the non-semantic
motivation of de-transitivizing alternations. As noted above, we can explain the
existence of intransitivizing alternations simply by referring to the need to
distinguish different non-linguistic events from each other. The differences may be
of various kinds, which increases the number of possible transitivity alternations
significantly. Thus, it is possible to distinguish different alternations functionally,
which is also reflected in the structure of the alternations in many cases. For
example, a decrease in the number of participants typically coincides with a change
in the number of arguments (even if there are exceptions). On the other hand, it is
not possible to classify de-transitivizing alternations in a similar way, since they
have the same conditioning factor. The only relevant distinction can be based on the
nature of the omitted argument. In passives, the argument is Agent, whereas in
antipassives it is the Patient. The other alternations cannot (as regards the de-
transitivization) be separated from typical passives and antipassives on the basis of
this criterion. Only the degree of de-transitivization varies. Furthermore, one
important reason for focussing on the structure was to show that the basic structure
of languages makes a significant contribution to de-transitivization. Since de-
transitivization is not directly related to non-linguistic transitivity in any way, the
structural nature of de-transitivizing alternation varies greatly. The most obvious
manifestation of this is the uneven cross-linguistic distribution of morphological
passives and antipassives. Every language has some means to express both these
functions, but languages diverge in what kinds of structural consequence the
expression of these functions has. Hence, the basic structure of languages impinges
on the nature of de-transitivizing alternations.

5.3.1.2. Transitivizing alternations

In the previous section, we have dwelled on transitivity-decreasing alternations. In

the present section the focus will be on the exact opposite of these, i.e. alternations
 409

that increase transitivity. Similarly to minus-transitivizing alternations, the

alternations to be presented are subclassified based on whether they affect the
number of participants, individual transitivity features or the mere number of
arguments (the last type is the ‘transitivizing equivalent’ of de-transitivizing
alternations). Also here we begin our illustration by the most obvious cases, i.e. by
presenting alternations that increase the number of participants in events.

5.3.1.2.1. Alternations increasing the number of participants (‘genuine

transitivizers’)

The cross-linguistically most typical transitivizing alternation is likely illustrated by

causatives. Causativization directly affects the number of participants in events,
since an agent is introduced. Causatives can be subdivided on the basis of
transitivity of the original clauses. The increase can be from one to two (‘he dies’
÷ ‘the man kills him’) from two to three (‘he eats bread’ ÷ ‘the man makes him eat
bread’) or from three to four (‘he gave John a book’ ÷ ‘the man made him give
John a book’). The first type concerns us most, since the causativization coincides
with transitivization in this case. Other types are examined, if necessary or relevant.
Another relevant division is based on the directness of causation. In typical cases,
the causation is direct and the agent intentionally causes an event to happen. In
others, the causation is less direct and the transitivization less complete. Other
alternations to be discussed include applicatives of different kinds (a more detailed
definition follows). We begin by examining causatives. It has to be stressed that it
is not the goal of the following presentation to discuss the notion of causativity
exhaustively. Causativization comprises numerous facets, and only some of these
can be taken account of in the following presentation.
As noted above, the emphasis of the present section lies on causatives derived
from intransitive clauses. As a result of these, an agent is introduced to an originally
intransitive event, which transitivizes the event along with the employed
construction. Semantically (and also structurally), these kinds of causative illustrate
the mirror image of anticausatives. In the most typical cases, the introduced agent
is acting volitionally and directly targets its action at the entity to be affected. The
action by the agent immediately results in a change-of-state in the patient. The
410

temporal distance between the two ‘sub-events’ is very short and it is very difficult
to divide the event into two parts. Hence, a typical causative event is viewed as a
single event with two participants.112 The action is salient in the sense that it can be
observed by others as well (this implies that the agent is acting physically). This
coincides with what has been labelled as ‘direct causation’ by e.g. Comrie (1989
([1981]: 171ff), DeLancey (1984b: 182) and Song (2001: 275f), ‘agent causative’
by Talmy (1976: 45), ‘volitional causation’ by Croft (1991: 166ff) and
‘manipulative causation’ by Shibatani (1976: 31) and Song (2001: 277)113. Typical
examples include the following, cf.

(1193) yc-p-ec-yt’
it(COL I)-PVB-break-FIN
‘It broke’
(1194) yc-p c-s-ec-yt’
it(COL I)-PVB-I(COL III)-break-FIN
‘I broke it’ (Abkhaz, Hewitt 1989: 168)

(1195) wo misa sit-]

the woman die-PERF
‘The woman died’
(1196) harsa-n ] wo misa-yat] siat- ]
Harsha-ERG the woman-DAT kill-PERF
‘Harsha killed the woman’ (Newari, DeLancey 1984b: 194f)

(1197) aqe al’~-j

frozen.fish thaw-3SG
‘The frozen fish thawed’
(1198) tiõ šoromê ~qêlêk al’~-š-mêlê
this man frozen.fish.DO thaw-CAUS-3SG
‘This man thawed the frozen fish’ (Yukaghir, Maslova 1993: 280)

Events denoted in (1193), (1195) and (1197) are intransitive, since they involve a

112
This holds generally true for typical transitive events, as noted in 3.2.

113
Song distinguishes direct and indirect causation based on the temporal distance of events, whereas
the latter distinction is based on whether the agent is physically acting on the patient or not.
 411

patient participant only. In (1194), (1196) and (1198), an external agent has been
introduced, which completely transitivizes the given events. Structurally, the three
illustrated cases are somewhat different. Examples from Abkhaz (see also (595)-
(596) from Lezgian and (603)-(604) from Djaru) illustrate a non-derived (see
Haspelmath 1993: 91) causative alternation. The verb is equally (un)marked in both
(1193) and (1194), in addition to which the given events are semantically equally
possible. Consequently, it is difficult to definitely state whether we are here dealing
with a causative or an anticausative. These constructions are genuinely labile. In
(1195) and (1196), the causation is lexical. Hence, we do not have any
morphological evidence to argue for the markedness of either construction (cf. also
e.g. Comrie 1989: 168). The ergative marking of the Agent or the dative marking
of the Patient are not valid criteria, since they are integral parts of transitive clauses
of Newari (cf. (365)-(366) from Lezgian). In this particular case, however, we may
argue that (1196) represents an alternation, since ‘killing’ can be regarded as
semantically marked in comparison with ‘die’ (in the world we live in, people
usually die rather than are killed). Examples (1197) and (1198) illustrate a genuine
causative alternation, since the verb is morphologically more complex in (1198).
Consequently, it is obvious that (1198) is derived from (1197).
Examples above illustrate cases in which the participant in the causativized
event is a passive patient that undergoes a change-a-state beyond its control. The
inchoative (intransitive) and causative (transitive) events differ from each other only
in that in the latter the change-of-state does not occur spontaneously, but is caused
by an external agent. The other possibility is that the participant in the underlying
intransitive event is an agent, in which case the causativization also affects the role
of the participant in question, since the introduction of an additional agent deprives
the original agent of some agentive properties (this coincides with ‘directive
causation’ in the spirit of Shibatani (1976: 31ff)). The overall transitivity of the
event is clearly lower than in cases like (1193)-(1198), since the patient is less of a
patient than in the previously illustrated cases, cf.

(1199) m cca cahil-]

child walk-PERF
‘The child walked’
412

(1200) misa-n ] wo m cca-yat] cahi-k hal-]

woman-ERG the child-DAT walk-CAUS-PERF
‘The woman made the child walk’ (Newari, DeLancey 1984b: 194)

(1201) kupa punthi-rna warrayi

child separate-PARTIC AUX
‘The children separated’
(1202) karna-li kupa punthi-ipa-rna warrayi
man-ERG child separate-TR-PARTIC AUX
‘The man separated the children’ (Diyari, Austin 1997: 173)

(1203) adam koš-tu

man run-PAST
‘The man ran’
(1204) adam-v koš-tur-du-k
man-ACC run-CAUS-PAST-1PL
‘W e made the man run’ (Turkish, Zimmer 1976: 399)

(1205) mam c natcn cwa

I.NOM dance.PRES
‘I dance’
(1206) maaw c nætuna
I.ACC dance.PASS.PAST
‘I danced (for some external reason)’ (Sinhala, Gair 1998b: 68f)

Events denoted above differ from (1193)-(1198) in that the role of the causee is not
that of typical patient, but the given participant shares common features with agents.
The causee is actively partaking in an event instead of being a mere passive target.
The degree of agency is somewhat reduced due to a decreased degree of
volitionality. Despite the obvious semantic changes, clauses above are structurally
identical to the basic transitive clauses or other direct causatives in the given
languages (cf. (1195) and (1196) from Newari, (720) from Diyari and (539)-(540)
from Turkish). Examples (1205) and (1206) do not illustrate typical causatives, but
they are presented here as a curiosity. The sole participant of (1206) is semantically
very similar to ‘genuine causees’ of (1200), (1202) and (1204). It is not performing
the action denoted willingly, but is somehow forced to partake in it. Hence, the
agency is reduced also in this case. Even if the overall transitivity of (1199)-(1204)
 413

is somewhat lower than that of (1193)-(1198), we may also claim that the causatives
illustrated above are the only ‘true causatives’. This can be justified by referring to
the markedness of the events and the direction of derivation. As noted above, it is
semantically quite difficult to argue justly for the higher markedness of either
inchoative or causative events. For example, breaking of things is equally likely to
occur spontaneously as to be caused by an external agent (cf., however, Haspelmath
1993b: 103). In cases like (1199)-(1204), on the other hand, the causative clearly
illustrates a marked event type. This is likely due to the fact that especially humans
typically partake in events volitionally without being forced to do so. On the other
hand, inanimate things are not capable of this, which makes causatives rather natural
in this respect. In the latter kind of causative, the introduction of an external agent
can accordingly always be regarded as an increase in transitivity, while in cases like
(1193)-(1198), the alternation can be either causative or anticausative. Hence, the
status of (1199)-(1204) as genuine causative alternations is higher. Notice that this
argument is based only on the notion of markedness. The effect on transitivity is
more significant in the former case.
The more agentive nature of the causee in the latter kinds of case has the
consequence that the degree of ‘patientness’ can be manipulated. This is not
possible at all (or is possible only very marginally) in cases like (1193)-(1198)
resulting from the passiveness of the causee. In examples above, the causee is forced
to act and the degree of agency is lower than in the corresponding intransitive
events. The structure employed is highly transitive. In the following examples, the
degree of agency is higher and the transitivity of the event lower (cf. (1043)-(1048)
above), cf.

(1207) k ‚ namcha-ka talli-n-ta

DET man-NOM run-PRES-DECL
‘The man runs’
(1208) k ‚-ka k‚ namcha-r ‚l talli-ke ha-ycss-ta
3SG-NOM DET man-ACC run-CAUS do-PAST-DECL
‘He made the man run’
(1209) k ‚-ka k‚ namcha-eke talli-ke ha-ycss-ta
3SG-NOM DET man-DAT run-CAUS do-PAST-DECL
‘He let the man run’ (Korean, examples courtesy of Hak-Soo Kim)
414

(1210) ewe-sel-Ø kad’d’ak-u (miine-w) kool-ukan

Even-PL-NOM Kaddak-ACC (wine-ACC) drink-CAUS.NFUT.3PL
‘Evens made Kaddak drink (the wine)’
(1211) ewe-sel-Ø kad’d’ak-tu (miine-w) kool-ukan
Even-PL-NOM Kaddak-DAT (wine-ACC) drink-CAUS.NFUT.3PL
‘Evens made/let Kaddak drink (the wine)’ (Even, Malchukov 1995: 14) 114

(1212) maduhan-li-l "bu-g sv b-i8e–k’e-k’-er

neighbour-OBL-ERG father-ADESS bear/4 4-kill-CAUS-CAUS-PRET
‘The neighbour forced father to kill the bear’
(1213) maduhan-li-l "bu-g sv b-i8e-k’-er
neighbour-OBL-ERG father-ADESS bear/4 4-kill-CAUS-PRET
‘The neighbour made father kill the bear’ (Hunzib, Van den Berg 1995: 108)

(1214) avan aval-ai vara-c vai-tt-aan

he she-ACC come-INF place-pa-3SGM
‘He made/had her come/caused her to come’
(1215) avan aval-ai vara-c cey-tt-aan
he she-ACC come-INF make-pa-3SGM
‘He made/had her come/got/caused her to come’ (Tamil, Fedson 1985: 13)

(1216) no-wila na ana

3R-go NOM child
‘The child goes’
(1217) no-pa-wila te anabou i jambata na ana
3R-CAUS-go CORE father OBL jetty NOM child
‘The father send the child to the jetty’
(1218) ku-hepe-wila (na iaku) di ana
1SG-REQ-go (NOM 1SG) OBL child
‘I ask the child to go’ (Tukang Besi, Donohue 1999: 213, 217)

The last example of each pair represents a case in which the degree of agency
associated with the causee is higher and the overall transitivity of the event lower.
Korean and Even are languages in which the different instances of directive
causation are distinguished by manipulating the case marking of causee. Both in

114
The Patient appears in brackets, since we are not interested in this aspect of the clause, it is not
marked as optional by Malchukov.
 415

Korean and Even, the accusative (i.e. highly transitive) marking of the Patient
coincides with direct causation. This is expected, since the overall transitivity is
higher in (1208) and (1210). In Hunzib and Tamil, the distinction is based on the
verb morphology. In Hunzib, the function of the causative morpheme is to add a
causer to the event, as in (1213). The second causative adds the nuance ‘force
someone to do something’ (Van Den Berg: ibidem, cf. (471) from Wolof in which
the second ‘causative’ in cases like this is interpreted as applicative). In Tamil, the
causative morphemes used are different. Tukang Besi illustrates a language in which
changes in the causative morpheme are accompanied by changes in the grammatical
status of the Patient. The Patient is typical only in (1217). The Patient of (1218)
cannot be cross-referenced on the verb nor can it be relativized or promoted to the
subject of passive (Donohue 1997: 217).
Examples above illustrate various instances of directive causation. As
predicted, the difference between manipulative (physical) and directive (‘verbal’)
causation is also expressed explicitly in many languages. A couple of examples are
illustrated below, cf.

(1219) honánwa-t istocí-n hómpeyc-ís

male-NOM baby-OBL eat.DIRECT.CAUS:LGR-IND
‘The man is feeding the baby’
(1220) honánwa-t istocí-n hómp-ipeyc-ís
male-NOM baby-OBL eat-make:LGR-IND
‘The man is making the baby eat’ (Creek, Martin 2000: 397)

(1221) akáli dokó-mé énda dóko kuma-sá-py-á

man the-ERG woman the die-CAUS-NR.PAST-3SG
‘The man killed the woman’
(1222) akáli dokó-mé énda dóko wápaka pe-ná l-é-á
man the-ERG woman the W abag go-3SG say-IM.PAST-3SG
‘The man caused the woman to go to W abag’ (Enga, Foley 1986: 153f)

(1223) *k ‚-ka k‚ tol-eke ttclc ha-ycss-ta

3SG-NOM DET stone-DAT drop.PASS.CAUS do-PAST-DECL
(He made the stone fall/he let the stone fall) (Korean, example courtesy of Hak-Soo Kim)
416

(1224) tanniir-ai-k kotikka vai-tt-aal (*cey-tt-aal)

water-ACC boil.INF place-pa-3SGF (make-pa-3SF)
‘She boiled the water’ (Tamil, Fedson 1985: 14)

Examples from Creek and Enga illustrate nicely the relevant distinction. In (1219),
the causer is physically making the causee eat (e.g. by spooning food into the
causee’s mouth). No control is exercised by the causee and the two subevents cannot
be separated (which is one of the criteria of high transitivity). In (1220), the
causation is directive. In Creek, this distinction is expressed solely by changing the
causative morpheme. Examples from Enga illustrate the iconic relation of semantics
and morphosyntax of causatives. More direct, physical causation is encoded
morphologically, whereas less direct causation is expressed periphrastically (cf. e.g.
Comrie 1989: 172). The two events are inseparable in (1221), while (1222) rather
comprises two distinct events construed as one. Examples from Korean and Tamil
are illustrated here simply for the purpose of showing that the kind of semantically
conditioned variation presented above is possible only in the case of directive
manipulation. The lower structural transitivity of (1220) and (1222), is
understandable, since the more active involvement of the patient is inversely
proporitional to transitivity.
In all the cases presented so far, the causation can be considered direct and
volitional in the sense that the agent is acting volitionally and can be viewed as
directly responsible for the occurrence of events. The events result from the action
by an agent. It is irrelevant whether the causation is manipulative or directive as
long as the relation between the action by the agent and the occurrence of events is
direct. This also implies that the causing action is directly targeted at the causee. The
causation can be far less direct than this. In this case, an event that ultimately causes
a patient to be affected is not necessarily targeted at the patient. The affectedness is
accidental. The agent and patient do not need to be in any contact with each other
and can even be located at distinct places. Indirectly caused events can in many
cases be thought of as involving two completely distinct events that in a particular
case seem to stand in a cause-effect relation. This relation is crucially different from
typical ones in which the causing is obvious and salient (cf. Song 2001: 258). The
independence of events also implies that the temporal distance between them can
 417

be long. A couple of examples are illustrated and discussed in the following, cf.

(1225) manaw gaw matu hpe ja-san ai

MaNaw TOP MaTu OBJ CAUS-die DECL
‘MaNaw killed MaTu’
(1226) manaw gaw matu hpe san shang un ai
MaNaw TOP MaTu OBJ die CAUSE DECL
‘MaNaw caused MaTu to die’ (Jinghpaw, Maran & Clifton 1976: 445)

(1227) da pe-gosa ringe

3SGA CAUS 1-be.good 3SGO
‘He healed her’ (directly, with spiritual power)
(1228) da puna ringe gosa
3SGA CAUS 2 3SGO be.good
‘He (did something which indirectly) made her well’ (Buru, Dixon 2000: 69, cited from
Grimes 1991: 211)

(1229) if lep seu-d’

he bread dry(TRANS)-FIN
‘He dried the bread’
(1230) if lep e-gu-d’
he bread dry(INTR)-CAUS-FIN
‘He caused the bread to dry’ (Nivkh, Comrie 1989: 172)

(1231) kotúé’ k’é lán ©we

liquor k’é died
‘S/he died from liquor’
(1232) kotúé’ xot’e lán ©we
liquor because.of died
‘S/he died because of liquor’ (Hare, DeLancey 1984b: 192)

(1233) 'adhaka (zayd-un hind-an)

laugh.CAUS.PERF.3SG.M. (Zayd-NOM Hind-ACC)
‘Zayd caused Hind to laugh’
(1234) dahikat (li-zayd-in)
laugh.PERF.3SG.F (because of-Zayd-GEN)
‘She laughed (because of Zayd)’ (Arabic, Premper 1988: 30f)
418

(1235) yobeto babeo na-hoti-ka

name.MASC paper.FEM CAUS-have holes-DECL.MASC
‘Yobeto made holes in the paper’
(1236) babeo hoti-ke (Yobeto ehene)
paper.FEM have holes-DECL.FEM (name.MASC due.to.MASC)
‘The paper has holes (due to Yobeto)’ (Jarawara, Dixon & Aikhenvald 1997: 82)

Examples (1225)-(1236) display both structural and semantic differences. What

these examples have in common is that an external instigator is introduced to an
intransitive event. The former example of each pair is a basic transitive clause. The
causation is less direct in the latter clauses, which manifests itself
morphosyntactically. Examples from Jinghpaw and Buru illustrate cases in which
direct causation is expressed morphologically, whereas periphrastic causativization
indicates a less direct causation. In Nivkh, the mechanisms of morphological
causativization vary according to the directness of causation. In these three
languages the marking of arguments remains constant. In Hare (cf. (1072) above),
Arabic and Jarawara, on the other hand, both the verb morphology and the argument
marking are affected. These languages have in common that the causative (or
transitive) verb morphology coincides with direct causation. Intransitive (non-
causative) verb morphology implies indirect causation. The difference to typical
intransitive clauses lies in the number of arguments. The mechanisms of
causativization are radically different depending on the directness of the
causation.115 Also the semantics of the examples concern us here.116 Example (1226)
is appropriate, for example, in cases in which someone hires an assassin to kill
MaTu or in which MaNaw fails to come to his aid (Maran & Clifton 1976: 446).
The ‘agent’ referred to is not the direct cause of the MaTu’s death, i.e. he does not
carry out the action that eventually kills MaTu. Example (1228) illustrates a typical
example of a case in which a possibly volitional and controlled action happens to
result in something not planned. The primary target of the action does not have to
be the patient. Examples from Nivkh are semantically rather similar to those from

115
It is doubtful, whether we can speak of true causativization in the latter cases.

116
The analyses are largely from the authors cited above.
 419

Jinghpaw. Example (1229) is appropriate, if someone purposefully dries bread by,

for example, putting in the oven. (1230), on the other hand, implies that someone
caused the bread to dry by forgetting to cover ir properly. Examples from Hare are
very interesting. In Hare, ‘agents’ like ‘liquor’ can appear both in transitive (cf.
(1072)) as well as intransitive frames, as in (1231).117 Example (1231) denotes an
event in which the liquor is directly responsible for a death.118 This means that the
person in question died as a result of immense alcohol consumption. The cause-
effect relation is salient. (1232), on the other hand, is more appropriate in case a
drunken driver kills someone. The driver is a genuine agent and the liquor can be
held only partly responsible. The causation is very far from being direct in any view,
since (1232) can be used, even if the person who dies has never tasted any liquor.
Drinking liquor and killing are clearly distinct events. Examples from Arabic
exemplify the difference between ‘concrete’ and ‘non-concrete’ causation.119 In
(1233), Zayd is purposefully trying to make Hind laugh, for example, by telling
jokes or by tickling her. In (1234), ‘she’ can laugh even if Zayd is not present. For
example, thinking about Zayd or something he has done can make someone laugh.
Examples from Jarawara are semantically rather close to each other (Dixon &
Aikhenvald: ibidem). However, the directness of causation makes a contribution
also here. (1235) is probably more appropriate, if Yobeto purposefully makes holes
to a piece of paper. In (1236), on the other hand, the same piece of paper can have
holes due to carelessness of Yobeto. Yobeto may have forgotten it lying close to a
fire place, in which case some parts may have caught fire, and, as a result, the paper
has holes. The causation is very indirect and also the person who lit the fire can be
responsible for the event.
In (1225)-(1236), we examined some instances of indirect causation. This
means that the instigator is not directly targeting its action at the patient. In many
cases, the actual agent is different from the one held indirectly responsible and the

117
This marking is not possible in the case of animate agents.

118
In case the agent is an inanimate entity the causation is always somehow indirect. These kinds of
case are not discussed in any detail here, since they were discussed in greater detail above.

119
Terms are our own.
420

cause-effect relation is rather conceptualized instead of real. For example, in (1226)

the actual agent is rather the ‘true killer’ (not referred to) than MaNaw. MaNaw is,
however, conceived of as being responsible for the event. Indirectness of causation
is in many cases related to a lower degree of intentionality. For example, in (1230),
it is very likely that the person responsible for drying the bread did not mean to
leave the bread uncovered. Notice, however, that indirectness of causation and
intentionality are not mutually exclusive. In (1230), it is also possible that someone
purposefully ‘forgot’ to cover the bread, since s/he did not want to eat it and
therefore made it inedible by causing it to dry. The causation is very indirect, even
if the action of the agent is intentional. In addition to indirectly caused events, there
are also genuine example of cases in which the causation is completely unintentional
or accidental. A couple of examples of this have already been discussed above (see
(1112)-(1121)). Below, we examine some examples that obviously illustrate distinct
causativizing mechanisms (this means that both of the examples are clearly derived
from intransitive clauses and that the accidental construction is not structurally
secondary to the typical one in this respect), cf.

(1237) k cc p-háan tráak

3SG.M CAUS 1-die buffalo
‘He slaughtered the buffalo’
(1238) k cc tòk háan múuc
3SG.M CAUS 2 die ant
‘He happened to kill an ant’ (Kammu, Svantesson 1983: 106)

(1239) |nh ôp d |q khlâyh

1SG CAUS 2 trap escape
‘I made the trap spring (on purpose)’
(1240) |nh ôp d |q ta-khlâyh
1SG CAUS 2 trap CAUS 1-escape
‘I made the trap spring’ (accidentally) (Chrau, Dixon 2000:70, cited from Thomas 1969: 100)

(1241) daikina-pe h «da-pada-pidana di-÷ahmeta di- a-nhi

afternoon-PL every-AFX-REM.PAST.INFR 3SG.NF-fear.CAUS 3SG.NF.stay-IMPF
‘He used to frighten them (villagers) every afternoon (with his cries)’
(1242) t•inu nu-na ha ÷ameta di-kwisa-ka
dog 1SG-OBJ fear.CAUS 3SG.NF-bark-REC.PAST.VIS
 421

‘The dog scared me, it barked’ (Tariana, Aikhenvald 2000: 156)

(1243) ngayu nhanu minha gundil bulii=ma-ni

1SG.NOM 2SG.GEN.ABS meat.ABS egg.ABS fall=CAUS-PAST
‘I dropped (lit. caused to fall) your egg’
(1244) nhanu minha gundil buli (ngadhun.gal)
2SG.GEN.ABS meat.ABS egg.ABS fall.PAST (1SG.ADESS)
‘I dropped your egg by accident/your egg fell’ (Guugu Yimidhirr, Haviland 1979: 125)

Examples (1237)-(1244) differ from (1225)-(1236), since the differences in the

marking follow from whether the causation is regarded as purposeful or accidental
(unintentional) (cf. also (365)-(367) from Lezgian). In the latter examples of each
pair, the causation can be viewed as accidental. This means that the agent has not
intended to instigate the given event. The agent is an agent involuntarily (cf.
Haspelmath 1993: 291). The agent may be intentionally and purposefully involved
in another action, but the extension to other, accidentally instigated events, is
unintentional. A typical example is provided by accidental dropping of things that
can, e.g., be due to walking by an object that happens to fall. Accidental causation
differs from indirect in that in accidental causation, the agent directly causes the
event to happen, which implies that it is also the actual agent in the event. In
distinction from typical transitive events, the intentionality is lacking. Languages
illustrated above diverge in how they distinguish between purposeful and accidental
causation. In Kammu and Chrau, these distinction is based only on the verb
morphology. In Kammu, morphological and periphrastic causation coincide with
typical and less typical causation, respectively. Chrau is rather similar to Kammu,
since the less typical causation is structurally more complex. In Chrau, the causation
is periphrastic in the case of purposeful causation, whereas accidental causation is
expressed both periphrastically and morphologically. In Tariana, intentionality
related to causation is expressed by attaching the appropriate person affix to the
causative verb, which is lacking in clauses denoting unintentional causation. Guugu
Yimidhirr is a combination of these two types, since the given semantic changes are
signalled both by manipulating the verb morphology and the case marking of Agent.
Causative verb morphology is restricted to typical causation.
Above, we have illustrated and discussed different causative alternations. The
422

illustration has been rather superficial and has ignored many aspects. This is simply
due to saving space. We believe, however, that the facets relevant for our purposes
have been examined sufficiently. All the types discussed have in common that the
causation (understood merely as an introduction of an external agent to an
intransitive event) correlates directly with transitivization (the only exceptions being
(1210)-(1213) and (1219)-(1220)). This means that causativized clauses are
obviously more transitive than the underlying intransitive ones. The types illustrated
diverge in the degree of transitivization, as shown. The kinds of causative illustrated
and discussed above are here labelled as ‘true causatives’, since, as noted
previously, causativization correlates directly with transitivization. The central
property of true causatives is the increase is from one to two participants. Moreover,
the introduced participant must be an agent (understood in a broad sense). The other
cross-linguistically frequent causative type is presented by causatives of transitives.
As was discussed above (see 5.2.4.), these do not exemplify genuine
transitivizations, since the introduction of the agent (or rather a causer) does not
coincide with an increase in the overall transitivity of events. The underlying event
involves an agent and it is, hence, not possible to introduce these features. It is rather
the case that the transitivity of the underlying event reduces, since the original agent
is deprived of complete agency. Since we have dwelled on the semantics (5.2.4.) and
structure (5.2.6.) of these alternations above, they will not be discussed here. We
also believe that a detailed study of transitive causatives would not reveal anything
that is not inferable from the typology proposed above.
Since our presentation has focussed on causatives, we have been illustrating
exclusively cases in which an agent are added to originally intransitive events. In
what follows, we will illustrate cases in which the participant introduced to an
intransitive event is a patient. We can distinguish between three types labelled as
extended intransitives120, benefactive/malefactive constructions and cognate object
constructions. The first type will further be subclassified. The illustrated types differ
crucially from each other in what kinds of consequence they have for the transitivity
of events. What these types have in common is that the resulting structures are
typically less transitive than typical transitive events (this is also partly reflected

120
Defined primarily semantically, i.e. differently from Dixon and Aikhenvald 2000b.
 423

morphosyntactically), but they naturally outrank intransitive events in transitivity.

In the following, each type is illustrated and discussed in the order they are
mentioned above.
As extended intransitives are here labelled intransitive events that are
extended to other participants. Typical examples of this are provided by cases like
‘he ran’ vs. ‘he chased the cat’ (these correspond roughly to what has been labelled
as S/A labile verbs or ‘agentive’ ambitransitives (see Mithun 2000: 86)). In this
particular case, an intransitive event of running is extended to an external
participant, which transitivizes the given event. Extended intransitives are here
subclassified on the basis of the semantic role of the sole participant involved in the
original event. Moreover, the degree of integration of the introduced participant into
the event structure is important. Both these aspects are related to linguistic
transitivity, as will be shown.
The first distinction between different types of extended intransitives is based
on the semantic role of the sole participant in the ‘extending event’. Examples are
illustrated below, cf.

(1245) wit-é nge-bruk-i omah

tree-DEF collapse-object=subject-goal=complement house
‘The tree fell on a house’ (Javanese, Poedjosoedramo 1986: 25)

(1246) talara kuda-yi ( õaliõu)

rain.ABS fall-PRES (1DL.EXCL.LOC)
‘It is raining (on us)’
(1247) talara-li õalina kuda-lka-yi
rain-ERG 1DL.EXCL.O fall-TR-PRES
‘The rain is pouring on us’ (Diyari, Austin 1981: 158)

(1248) kaie vutuhutu

3SG 3SG.real.REDUP.abuse
‘He swore’
(1249) kaie vuti taataa onene
3SG 3SG.real.abuse father poss.man.3SG
‘He abused his father’ (Paamese, Crowley 1982: 154)
424

The example from Javanese illustrates an extension of a patientive event (of

‘falling’) to another participant. The role of the instigator in (1245) is twofold. It is
a patient in the preceding event, whereas it is conceived of as an agent in (1245).
Hence, events like (1245) involve two events and their semantic transitivity is
clearly lower than that of typical transitive ones. Examples (1246) and (1247) can
be regarded as an intermediate form between cases like (1245), in which the sole
participant is a patient, and (1248) and (1249) that illustrate an extension of an
agentive event to another participant. In (1246) and (1247), the extended event is
autonomous and the sole participant is neither a true agent nor a true patient.
Similarly to (1245), the extension cannot be volitional or controlled. On the other
hand, the event in question occurs without any external causation, which makes it
more similar to (1248) and (1249). Examples (1248) and (1249) illustrate a case in
which an agentive event of ‘swearing’ is targeted at another participant that gets
affected by the event. Structurally, the Paamese examples are rather interesting,
since the intransitive verb is explicitly marked. We are here dealing with an obvious
transitivization, though, since the verb morphology is transitive in (1249). Examples
(1245), (1247) and (1249) can all be regarded as transitive events, since they involve
two participants as a result of the kinds of extension presented. The events are rather
similar to each other in their patientive features, since they all have an affected
patient. Their overall transitivity, however, diverges due to obvious differences in
agency. (1249) can be regarded as more transitive than (1245) and (1247), since the
instigator of the event is a typical agent.
Another relevant distinction is based on how intimately the introduced
participant is integrated into the semantics of the resulting event. As regards high
transitivity in general, both participants are integral parts of typical transitive events.
They both contribute to transitivity (and nature) of events. Extended intransitives are
readily dividable into two depending on whether the resulting events are similar to
transitive events in this respect or not, cf.

(1250) pu 8’-Ø ti-§imlk-tx ( §u »-»mi»)

come-he -man- (PREP-us)
‘The man came towards us’
 425

(1251) pu 8’-m-tu »s ti-§imlk-tx

come-m-he/us -man-
‘The man attacked us’ (Bella Coola, Davis & Saunders 1997: 64)

(1252) karna wapa-yi (wilha-nhi)

man go-PRES (woman-LOC)
‘The man is going with the woman’
(1253) karna-li wilha wapa-lka-yi
man-ERG woman go-TR-PRES
‘The man takes the woman’ (Diyari, Austin 1997: 172f)

(1254) hän ajo-i (*vei) paimio-on

he.NOM drive.PAST-3SG (took) Paimio-ILL
‘He drove to Paimio’
(1255) hän ajo-i / vei minu-t paimio-on
he.NOM drive.PAST-3SG / take.PAST.3SG 1SG-ACC Paimio-ILL
‘He drove/took me to Paimio’ (Finnish)

(1256) hän sylkäi-si (minu-a)

he.NOM spit-3SG.PAST (1SG-PART)
‘S/he spat (at me)’ (Finnish)

(1257) ¨ali-di rasul- ’u (tata) tudi

Ali-ERG Rasul-CONTESS (saliva) spit.PAST
‘Ali spat (saliva) at Rasul’ (Godoberi, Kibrik 1996: 144)

(1258) yura-Ø wa õka-ku

man-NOM speak-NARR
‘Man spoke’
(1259) yura-õa atu-Ø wa õu-õu-ku
man-ERG woman-NOM say-TR-NARR
‘The man spoke to woman’ (Atjnjamathanha, Schebeck 1976b: 543)

(1260) rlke werne-me

wind blow-NONP
‘The wind is blowing’
(1261) rlke-le pipe mape werne-me
wind-ERG paper PL.O blow-NONP
‘The wind is blowing the papers (around)’ (Aranda, W ilkins 1989: 224)
426

(1262) mài-ma-ki-a-danya-ka uda

come-EMP-MOOD-MOOD-3PL.CONT-PERF EMP.3PL
‘They just came’ (i.e. direction/purpose not important)
(1263) pa-mài-ng-ma-ki-a-danya-ka uda
CAUS-come-ng-EMP-MOOD-MOOD-3PL.CONT-PERF EMP.3PL
‘They just came’ (to stay, i.e. they just moved here) (Kambera, Klamer 1998: 182)

In (1250)-(1255), the patient is intimately integrated into the resulting event, while
in (1256)-(1261) the integration is less complete. The two types of extended
intransitive differ from each other also in the kind of effect the introduction of the
patient has for the general nature of events. In (1250)-(1255), the effect is more
drastic and the two instances of the basically same event are rather different in
nature. The patient is an integral part of the extended intransitive event and the event
cannot occur without there being a patient present. For example, ‘attacking’ as a
transitive event is clearly different from an intransitive act of ‘coming towards
someone’. ‘Attacking’ requires that there be an entity present targeted by the action.
In (1256)-(1263), on the other hand, the effect of patient on the nature of the event
is less significant and we are rather dealing with two different instances of one
event. For example, for the nature of spitting it is irrelevant whether the action is
(purposefully or accidentally) targeted at another entity or not. The patient can be
omitted without this having any consequences for the basic semantics of the event.
Transitivization is less complete in the latter cases. It is noteworthy that the degree
of transitivization does not necessarily correlate with the degree of patient
affectedness, but patients in (1256)-(1261) can be more directly affected than those
in (1250)-(1255). The languages exemplified differ from each other as regards the
morphosyntactic effects the extension has on the clause structure. As the examples
show, the rather obvious semantic differences are not necessarily reflected
morphosyntactically. Only in Finnish and Godoberi, the latter kind of construction
is structurally less transitive.121 Languages above can be subdivided based on
whether the changes are accompanied by a morphological modification of the verb

121
W e can say next to nothing about the typology of these kinds of construction, since grammars
rarely analyze extended intransitives of this kind in any detail. The topic thus merits an independent,
detailed study of its own.
 427

or not. In Finnish, Godoberi and Aranda, the verb morphology remains constant
despite the introduction of a patient, whereas in Bella Coola, Diyari, Atjnjamathanha
and Kambera, the verb is transitivized. In Finnish, the extension illustrated in (1254)
and (1255) can be expressed both non-morphologically and lexically. What is
important for our purposes is that viedä (‘take’) cannot appear in an intransitive
frame, but the extension necessarily transitivizes the intransitive ajaa. Hence, the
introduced patient is an integral part of the semantics of the resulting event.
As regards the semantics of the illustrated cases, the two types are readily
subdividable into two. The first type comprises cases in which an intransitive spatial
action is extended to another participant. The entity involved in the original event
is an agent and the extension is purposeful. Examples from Diyari illustrate another
important facet of transitivity. In (1252), there are two participants (one of which is
optional) involved in the same intransitive activity. Hence, both of them are agents.
In (1253), on the other hand, the relations obtaining between the participants are
clearly different. Only ‘man’ qualifies as an agent, whereas the role of ‘woman’ has
shifted from agent to patient. Consequently, the semantic role assignment in (1253)
is typical of transitive events. The other type of extended intransitives is exemplified
by cases in which a bodily action (‘spitting’, ‘vomiting’, ‘urinating’, ‘speaking’ etc.)
is extended to another participant or in which a force of nature (‘fire’, ‘wind’ etc.)
affects another participant. In these cases, the extension does not have to be
purposeful and the target is always an optional part of the resulting events (i.e. it
does not affect the basic nature of events). Examples from Kambera do not readily
illustrate neither of the types. The difference between (1262) and (1263) lies in the
fact that in the latter case the given activity has a definite goal that is lacking in
(1262). What is of interest in the present context is that in the latter case the verb is
causativized (cf. also (201)).
We have illustrated causatives and extended intransitives as clearly distinct
alternation types. Despite obvious differences, they can also be thought of as two
different realizations of one basic alternation type in some cases. In this view, these
alternation types differ from each other only in the point-of-view chosen.
Semantically, extended intransitives are then regarded as a subtype of causatives.
This view is justified, since the resulting structure is (semantically) the same. For
example, in (1253) we are dealing with an extended intransitive, since the agent
428

(‘man’) is chosen as the viewpoint. Since the given event already involves an agent,
the introduced participant is a patient by default. If the only participant in (1252) is
a patient, then (1253) is a causative.122 On the other hand, however, this analysis is
not applicable to cases like (1256) and (1257) in which a causative reading is
excluded. One cannot be spat at without there being an agent responsible for the
spitting. This distinction correlates roughly with the division into the two types
illustrated above. Possible problems arise in cases like (1258)-(1259). These cases
are more similar to (1256) and (1257), but there are also cases in which both
readings are possible (e.g. ‘the wind is blowing’ vs. ‘the door shut’ ÷ ‘the wind shut
the door’).
In extended intransitives, an affected patient is introduced to an intransitive
event. The action in question is directly (yet perhaps involuntarily) targeted at the
affected participant that is also the primary target of the action. In this respect,
extended intransitives are similar to typical transitive events. The second type of
alternation that introduces an affected participant comprises cases in which the
added participant is not directly targeted by the action despite being indirectly
affected by it. Alternations at issue include malefactive and benefactive alternations,
cf.

(1264) bujun-Ø (töör-duk) il-ra-n

reindeer-NOM (ground-ABL) stand.up-NFUT-3SG
‘The wild reindeer stood up (from the ground)’

(1265) bujusem õe-Ø buju-m ila-w-ra-n

hunter-NOM reindeer-ACC stand.up-ADVERS-NFUT-3SG
‘The wild reindeer stood up, the hunter was negatively affected’ (Even, Malchukov 1995: 22)

(1266) rul=ni§ ka-§in= §a § §a-ka-lu §-hno §

snake=ERG 1SG.POSS-house=LOC 3SG.SUBJ-1SG.OBJ-enter-MAL
‘A snake came into my house on me’ (Lai, Peterson 1998: 99)

122
This is only a theoretical possibility, since causative and applicative are distinct alternations in
Diyari.
 429

(1267) ùbúr á-p āār`

Ubur CMPL-jump
‘Ubur jumped’
(1268) dháag ] á-p ārr`-ì ùbúrr-ì
woman CMPL-jump.BEN-SUF Ubur-ERG
‘Ubur jumped for the woman’ (Päri, Andersen 1988: 304)

(1269) mein hund ist (mir) gestorben

my.NOM dog be.PRES.3SG (1SG.DAT) die.PARTIC
‘My dog has died (on me)’ (German)

(1270) isä laulo-i (poja-lle-en)

father.NOM sing.PAST-3SG (son-ALL-3POSS)
‘The father sang (for his son)’ (Finnish)

(1271) umukoôbwa a-ra-som-er-a umuhuûngu igitabo

girl she-PRES-read-BEN-ASP boy book
‘The girl is reading a book for the boy’ (Kinyarwanda, Kimenyi 1980: 32)

(1272) Ø-q’a-y- c-w ck’c-n-w-s

3-HOR-3-N PRES-kill-FUT-DEF-AFF
‘(S)he will kill him/her (at some definite time)’
(1273) Ø-q’a-y- c-w ck’c-a-n-w-s
3-HOR-3-N PRES-kill-ACC-FUT-DEF-AFF
‘(S)he will kill him/her (kin of the speaker) (at some definite time)’ (Kabardian, Colarusso
1989: 316)

(1274) yara ya-ka-ta õ-kra-t-akn

tree V PL V PL T-1SG A-COM(APPL)-cut-PERF-3SG D
‘I cut his trees for him’ OR ‘I cut his trees’ (negative effect on him)
(1275) yara ya-ka-kra-õa-t-akn
tree V PL V PL T-1SG A-BEN(APPL)-cut-PERF-3SG D
‘I cut the trees for him’ (Yimas, Foley 1991: 310)

In all the cases exemplified above, an indirectly affected participant is introduced.

This does not have any (major) effect on the basic transitivity of events at issue.
Examples above differ from each other in whether the participant in question is
positively or adversely affected. There is also variation in whether the added
430

participant can be regarded as the primary (yet not direct) target of the activity. The
latter distinction is sensitive to transitivity of the underlying events. If the original
event is intransitive, the indirectly affected participant is by default the primary
‘target’ of the event, even if the activity is not directly targeted at the given entity.
This implies that the affectedness is not physical, but the given participant is
somehow mentally affected, which also implies that the degree of affectedness is
somewhat lower than in typical transitive events. Hence, both the degree and the
nature of affectedness distinguish bene/malefactive alternations from typical
transitive events. The contribution of indirectly affected participants to the structure
of events is radically different, if the underlying event is transitive (exx. (1271)-
(1275)). In this case, the event in question already involves a directly affected
participant that coincides with the primary target of the action. Hence, the
introduced participant cannot be the primary target. Events denoted in (1271)-(1275)
involve therefore both a directly and an indirectly affected participant. The basic
nature of the transitive (two-actant) event is not affected in any way as a result of the
introduction of new participants (cf. discussion in 5.2.4.). The action of the agent
is targeted at the primary patient and the transitive event as a whole affects the
introduced participant. In addition to the semantic differences, examples above
differ from each other as regards the structural consequences the introduction of
indirectly affected participants has. Examples (1264)-(1268) exemplify complete
structural transitivizations.123 The case frame employed is the same as in basic
transitive clauses. In Even, the introduced affected participant occupies the Agent
slot, whereas in Lai and Päri, it is morphologically a Patient. Examples (1269)-
(1270) present cases in which lower semantic transitivity correlates with less
transitive morphosyntax. In German and Finnish, the introduction of malefactives
or benefactives is not (necessarily) accompanied by changes in the verb morphology.
Examples from Finnish and German also illustrate the differences between
patientive (German) and agentive (Finnish) events. Examples (1271)-(1275)
illustrate the introduction of indirectly affected participants into transitive events.
Example (1271) from Kinyarwanda represents a very typical benefactive applicative
of Bantu languages. The benefactive is added to the clause core. In Kabardian, the

123
There are also structurally less transitive malefactive alternations is Even, see Malchukov:ibidem.
 431

introduction of an indirectly affected participant is signalled by attaching an

accusative affix to the predicate. Examples from Yimas, for their part, illustrate two
morphologically distinct bene/malefactive (glossed as comitative and benefactive
by Foley) affixes. These differ from each other in two respects. First (as the
translations show), (1274) can be interpreted as both benefactive and malefactive.
Only the former is possible in (1275). Second, the two cases are different as regards
the presence vs. absence of the indirectly affected participant. The more natural
interpretation of (1274) is that the indirectly affected participant is present during
the event. In (1275), this does not have to be the case.
The two previously illustrated alternation types introduce an external
participant to events. These types differ from each other in that in the former case
the added participant is a directly affected patient, whereas in the latter the effect on
the given participant is indirect. The last type of ‘Patient adding alternation’ is
represented by cognate object constructions. In the present context, the label is used
in a somewhat unorthodox way. Typically, the term refers to constructions like he
dreamed a terrible dream last night in ‘which the semantic content of the object is
more or less identical to that of the verb which governs it’ (Trask 1993:48). It seems
that in the most typical cases the lexical form of the object is derivable from the verb
(e.g. dream a dream). These are naturally also here taken into account. In addition,
cases in which the structural derivation is far less obvious are taken account of, if
the semantic condition is met. The semantics of the verb and that of the object does
not have to be identical, but the object must not, however, add anything to the
meaning that does not inhere in the semantics of the verb. Hence, the alternation
does not add anything ‘external’ to the underlying intransitive event. The difference
between cognate object constructions and extended intransitives is best illustrated
in light of differences between the likes of ‘he spoke to me’ vs. ‘he speaks Muna’.
In the former case, the target is external to the act of speaking, whereas in the latter
the ‘target’ inheres in the semantics of speaking. Typically, cognate objects refer to
‘non-concrete products’ of intransitive events that exist only as long as the event
proceeds. Hence, transitivization is not concrete in the sense that a new, concrete
participant is introduced. The primary function of the explicit object expression is
to specify the nature of events along with the nature of patients. The function is thus,
in a way, the opposite of de-transitivizing alternations illustrated in (1179)-(1188),
432

which entitles us to label them as transitivizing. A couple of structurally different

cognate object constructions are examined below, cf.

(1276) w Ç manu-na ca hil-a

the man-ERG walk walk-PERF
‘The man walked (a walk)’ (Newari, Givón 1995: 91)

(1277) ¨ali(-di) (šanša) šami

Ali-ERG (whistle) whistle.PAST
‘Ali whistled/blew a whistle’ (Godoberi, Kibrik 1996: 117)

(1278) isä laulo-i joululaulu-n

father.NOM sing-3SG.PAST Christmas.carol-ACC
‘The father sang a Christmas carol’ (Finnish)

(1279) ayenge arrente-Ø angke-me

1SG.S Arrente-NOM speak-NP
‘I speak Arrente’ (Aranda, W ilkins 1989: 167)

(1280) un-na õasa cho õs-u-n-na

he-OBL(ERG?) fish sell-3U-NEG-NML
‘He does not sell fish’
(1281) khan athpare riõ ocetnu-lok a-cek-na
you Athpare language nice-COM 2-speak-NM L
‘You speak the Athpare language well’ (Athpare, Ebert 1997: 123)

(1282) mawun `aru man-an

man.ABS Djaru talk-PRES
‘The man talks Djaru’ (Djaru, Tsunoda 1995: 97)

Newari, Godoberi and Finnish exemplify languages in which the morphosyntactic

status of cognate objects is identical to typical Patients. Hence, the resulting
structure is typically transitive. Examples (1279)-(1282) illustrate cases in which the
lower semantic transitivity of cognate object constructions manifests itself
morphosyntactically. Notice that all the examples that display lower structural
transitivity describe the ability to speak a language (cf. also (630) from Diyari).
Example (1280) does not exemplify a cognate object construction, but it is given
 433

here in order to explicitly illustrate the lower structural transitivity of cognate object
constructions in Athpare.

5.3.1.2.2. Alternations increasing the number of arguments (‘syntactic

transitivizers’)

Above, we have illustrated and discussed various alternations that increase the
number of (core) participants partaking in events. The resulting structure is typically
more transitive than the underlying one (this is the case when these alternations are
applied to intransitive clauses). Since typical transitive events involve an agent
(instigator) and a patient (affected participant), alternations that introduce either of
these naturally increases the transitivity of events (and clauses). All other kinds of
alternation should be excluded from the notion of genuine transitivizing
alternations. However, as will be shown (and is generally known), there are
languages in which also the introduction of peripheral participants produces an
obvious structural transitivization. Below, we will briefly examine these. The
presentation is rather cursory, since numerous examples have already been
illustrated above.124
In the previous section, certain kinds of alternation have been included in a
specific type primarily on the basis of semantic criteria. This approach is the most
appropriate one, since the participants introduced to events (and clauses) make a
significant contribution to semantic transitivity, which entitles us to include them
in the discussion despite the possible irrelevance for structural transitivity. In what
follows, the emphasis lies on structural features. This means that arguments have
to be added to the clause core in order that we can speak of a genuine transitivity
alternation. Semantics and structure disagree in many of these cases. This approach
is appropriate here, since it enables us to focus on relevant cases. Furthermore, since
participants introduced do not make any contribution to semantic transitivity, we can
only use strutural criteria to judge whether we are dealing with a true alternation or
not. For example, cases like (1283)-(1284) are not regarded as genuine transitivity
alternations, whereas (1285)-(1294) illustrate typical instances of applicatives, cf.

124
For a more detailed discussion of applicatives, see Peterson 1998.
434

(1283) isä juoksee (poika-n-sa kanssa)

father.NOM run.PRES.3SG (boy-GEN-3POSS with)
‘The father is running (with his son)’ (Finnish)

(1284) der vater läuft (mit sein-em sohn)

ART.NOM father run.PRES.3SG (with his-DAT son)
‘The father is running (with his son)’ (German)

(1285) kamij c-n v hiteraw c ja-kãi-hu-v

1SG-ERG Hiteraw c 1SG-COM-walk-DYN
‘I am walking-with Hiteraw c’ (Yanomami, Dixon & Aikhenvald 1999b: 349)

(1286) ee a-h vng (hx ]]h go)

father FORM-go.downriver (canoe in)
‘Father goes downriver in a canoe’
(1287) hx ]]h ee ga-h vng
canoe father in-go.downriver
‘Father goes downriver in a canoe’ (lit. Father goes downriver-in a canoe) (Nadëb, Martins
& Martins 1999: 262)

(1288) pablo i-ata-si-ke-ri ariberito

Paul he-go-PURP-PAST-him Albert
‘Paul went with Albert in his mind’ (Nomatsiguenga, Payne 1997: 189, cited from W ise 1971)

(1289) umuhuûngu a-rá-ririimb-a (n’îshávu)

boy he-PRES-sing-ASP (with.sorrow)
‘The boy is singing with sorrow’
(1290) umuhuûngu a-rá-ririimb-an-a ishávu
boy he-PRES-sing-with(APPL)-ASP sorrow
‘The boy is singing with sorrow’ (Kinyarwanda, Kimenyi 1980: 42)

(1291) booi §a-ka-toon-ka §n

chief 3SG.S-1SG.O-meet-PRIOR
‘He met the chief ahead of/before me’
(1292) §a-law §a-ka-thlo §-taak
3SG.POSS-field 3SG.S-1SG.O-work-RELINQ
‘He left me and hoed his field’ (Haka Lai, Peterson 1998: 36f)
 435

(1293) bill co·ka-n ho·cceyc-is

Bill letter-OBL write.LGR-IND
‘Bill is writing a letter’
(1294) bill isho Acceycka co Aka-n is-ho Acceyc-is
Bill pen letter-OBL INSTR.APPL.LGR-IND
‘Bill is writing a letter with a pen’ (Creek, Martin 2000: 390)

In (1283) and (1284), the peripherality of the introduced participant is manifested

in the status of arguments (or rather adjuncts). These are completely optional, in
addition to which the constituents in question appear in non-core case forms.
Furthermore (and perhaps more importantly), the introduction of these participants
into events does not have any effect whatsoever on the transitivity of clauses. Both
(1283) and (1284) involve two participants both of which are, however, involved
in the same intransitive activity, which means that neither of them can be regarded
as a targeted entity of any kind. Examples (1285)-(1294) are semantically similar to
(1283) and (1284), but the structural differences are obvious. In (1285)-(1294),
semantically peripheral participants are referred to by core arguments. The core
status is made possible by manipulating the verb morphology, which also makes the
expression of semantically peripheral arguments obligatory. In (1285)-(1292), which
are derived from intransitive clauses, the case frames of clauses are identical to basic
transitive clauses. In (1293) and (1294), an instrument is added to the clause core.
Above, we illustrated rather many examples of applicatives in order to give some
kind of overview of the variety of semantic roles associated with the applied objects
(see also (580)-(586) from Haka Lai).
Applicatives illustrated above are the most typical examples of transitivity
increasing alternations that do not affect event transitivity. Another type that could
be regarded as this kind of alternation is provided by causatives of transitives. As
noted above (see 5.2.4.), the introduction of a causer to a transitive event does not
increase semantic transitivity, but the effect is rather the opposite of this. The overall
transitivity of the event remains more or less the same, since the agentive features
that were originally properties of a single entity are divided between two entities.
Their number remains unchanged. The effect of causativization of transitive clauses
on the nature of events is, however, radically different from that caused by
applicatives, which also implies that their primarily structural nature is motivated
436

differently. The causer can be regarded as a core participant the introduction of

which affects the overall nature of events in a significant way. What entitles us to
regard them as primarily structural alternations is that they do not transitivize the
underlying transitive event. There can only be a certain amount of agentive features
involved in events, which makes it impossible to increase the number of transitivity
features.

5.3.1.2.3. Alternations increasing individual aspects of semantic transitivity

In cases illustrated above, the introduced participants or arguments are responsible

for transitivization. The last transitivizing alternation type discussed here comprises
cases in which the number of participants and arguments remains constant, and only
individual transitivity features are affected. These can be divided into three based
on whether they affect both relevant transitivity parameters or whether the primary
effect is on either of the relevant aspects.125 This type represents the opposite of the
alternations illustrated in 5.3.1.1.1.2.. The main difference between these two types
lies in the direction of derivation. In cases illustrated below, the starting point of a
derivation is less transitive than the resulting structure. In many cases, this coincides
with semantic markedness (cf. (365)-(367) and the discussion). Also the actual
distribution of these types is somewhat different. Each type will be exemplified in
what follows.
The first subtype of transitivizing alternation insensitive to the number of
participants or arguments is presented by alternations that affect both agency and
affectedness parameters. In this case, an event that is inherently relatively low in
transitivity is transitivized. A couple of examples are illustrated below, cf.

(1295) na fesili le leoleo i le tam ~loa

PAST ask ART police LD ART man
‘The police asked the man’
(1296) na fesili=gia e le leoleo le tam ~loa
PAST ask=ES ERG ART police ART man

125
In some cases it might be very difficult to say, whether the effect is only on either aspect, because
of which we have opted for using the label ‘primary effect’ in this context.
 437

‘The police questioned the man’ (Samoan, Mosel & Hovdhaugen 1992: 732)

(1297) nii ong-jo pehkoro iirong-ohna-na

1SG DEM.M-PURP boy get.angry-1S a.PRES.PROG-F
‘I am angry for the sake of this boy’
(1298) nii ong pehkoro iirong-ee-uhna-na
1SG DEM.M boy get.angry-APPL-3O.1A.PRES.PROG-F
‘I am angry with this boy’ (Motuna, Onishi 2000: 132)

(1299) mies rakasta-a nais-ta (*naise-n)

man.NOM love.PRES-3SG woman-PART (woman-ACC)
‘The man loves the woman’
(1300) mies rakasta-a naise-n kuoliaaksi
man.NOM love.PRES-3SG woman-ACC to.the.death
‘The man loves/will love the woman to death’ (Finnish)

(1301) *t ~ b| xi|o m ~o ài
3SG b| small cat love
‘S/he loves the kitten’
(1302) t~ b| xi|o m ~o ai de yao s0
3SG b | small cat love CSC want die
‘S/he loves the kitten so much s/he wants to die’ (Mandarin Chinese, Li &Thompson 1989:
466ff)

It is not justified to claim that only one of the transitivity parameters, but rather
transitivity as a whole has been affected in (1295)-(1302). All these examples have
in common that the action of the agent can be regarded as more intense, which also
increases the affectedness parameter. The event profiled in (1295) refers to a single
act of ‘asking’, whereas in (1296) the event consists of a series of questions. The
patient can also be viewed as more affected, since the questioning is necessarily
targeted at the ‘man’. Examples (1297) and (1298) are rather similar to (1295) and
(1296), since the ‘action’ of the agent is also here more intense. (1297) simply states
that the speaker is angry of because of the boy, whereas (1298) implies that the
agent is expressing his anger, which may have direct consequences for the patient.
Examples from Finnish and Mandarin illustrate two instances of ‘loving’. In (1299)-
(1302), the shift from less transitive to highly transitive constructions results in a
438

‘dynamicization’ of a formally stative relation. (1299) and (1301) merely state that
someone loves someone/something without any further implications. The
experiencer is not actively doing anything and the stimulus does not need to be
affected in any way. Examples (1300) and (1302), on the other hand, imply that the
experiencer is actively doing something to prove his/her love. The stimulus is rather
a patient directly affected by this.
Transitivizing alternations whose primary effect is on the affectedness
parameter can be divided into three subtypes. In the first case, the differences in
affectedness are most obvious. A single basic event can affect the patient to different
degrees. Examples include the following, cf.

(1303) kaie mutau (eni tirausis onene)

3SG 3SG.real-defecate (OBL trousers POSS.man.3SG)
‘He defecated (on his shorts)’
(1304) kaie mutau-ti tirausis onene
3SG 3SG.real.defecate.TRANS shorts POSS.man.3SG
‘He shat his shorts’ (Paamese, Crowley 1982: 191)

(1305) kua mohe a ia he fale

PERF sleep ABS he in house
‘He has slept in the house’
(1306) kua mohe e ia e timeni
PERF sleep ERG he ABS floor
‘He has slept on the floor’ (there is some effect on the patient) (Niuean, Seiter 1980: 63)

(1307) naa bùgee shì

1SG.PERF beat(II) 3SG.M.ACC
‘I struck him’
(1308) naa bugèe shì
1SG.PERF beat(IV) 3SG.M.ACC
‘I knocked him down’(Hausa, Heide 1989: 22)

(1309) na-÷uku-i-pidana naha it•ida-pe-ne

3PL-fall-CAUS-REM.PAST.INFR they turtle-PL-INSTR
‘They (devils) made (some woodchips) fall down with the help of turtles (axes)’
(1310) phia nuha panisi-nuku pi- a-bala
you I house-TOP.NON.A/S 2SG-hit-EVERYW HERE
 439

pi-÷uku-i-ta-ka
2SG-go.down-CAUS1-CAUS2-DECL
‘You destroyed my house completely (lit. hit everywhere - make come down) (said the evil
spirit to a man in his dream)’ (Tariana, Aikhenvald 2000: 158)

Examples above illustrate the opposite of (1031)-(1036), since the affectedness of

patient increases. In Paamese and Niuean, a non-core participant usually not
regarded as an integral part of events is considered one. Since the degree of
affectedness varies, we are dealing with a transitivizing alternation. Examples
(1303) and (1305) illustrate either an extended intransitive (Paamese) or a typical
intransitive construction with a locative phrase (Niuean). In (1304) and (1306), the
optional participant is viewed as being somehow directly affected by the given
event. The alternation does not affect the agency in any relevant way. Examples
from Hausa describe inherently transitive events. What distinguishes (1307) and
(1308) from each other is the degree of patient affectedness. In (1308), the patient
is more thoroughly affected. Tariana has two distinct causatives (i- vs. i-ta-) that can
be distinguished on the basis of the degree of patient affectedness. In (1310), the
affectedness is complete and the construction outranks (1309) in transitivity.
Another type of transitivizing alternation, which is caused by changes in the
affectedness parameter only, is illustrated by cases in which the patient can be
regarded as an incremental theme (cf. Dowty 1991, see also the discussion in
5.3.1.1.1.2.), cf.

(1311) a ngal“k a m “nga “r a ngik “l

child eat.IMPF.PRES PREP fish
‘The child ate the fish’
(1312) a ngal“k a kill-ii a ngik “l
child eat.PERF.PAST-3SG fish
‘The child ate the fish up’ (Palauan, Mosel 1991b: 404)

(1313) a õki-§a dzu-k-a

we.EXCL-ERG eat-1P-ERG
‘W e ate’
(1314) a õki-§a dzu-kh c§-k-a
we.EXCL-ERG eat-take.away-1P-ERG
440

‘W e ate it up’ (Dumi, Van Driem 1993: 211)

(1315) ngath ngaungu ithab koei puil pathamadhin

I.ERG myself these big tree.ABS.PL cut.PL.P.PAST
‘I cut down these big trees’
(1316) ngai ngaungu ithab koei puin pathaidhin
I.ABS myself these big tree.OBL cut-PAST
‘I cut down all these big trees’ (Kala Lagau Langgus, Comrie 1978:363, cited from Bani &
Klokeid 1976: 278)

In (1311)-(1316), the differences in affectedness stand in a close relation to the

degree of completedness of events (cf. (1025)-(1030) and the discussion). The
differences in affectedness are due to whether the patient is viewed as being
completely affected. For example, in (1311) the patient is directly affected, but the
event as a whole is less completed (and the patient consequently less completely
affected) than in (1310). In Dumi, the differences are expressed only by
manipulating the verb morphology. Examples from Kala Lagau Langgus are rather
interesting, since semantic and structural transitivity obviously disagree. The
translations suggest that (1316) is more transitive, since the patient is more affected
due to the total completion of the event. However, morphosyntactically, (1316) is
less transitive than (1315). Since the structural intransitivization is not motivated by
a decrease in transitivity, (1316) cannot be regarded as a typical intransitivization,
either. Because of the primacy of semantics, (1316) is here regarded as a
transitivizing alternation with a rather untypical structural realization.
Alternations illustrated in (1303)-(1316) follow from changes in the
affectedness of the primary target.126 Moreover, the affectedness of ‘secondary
targets’ can also be relevant, cf.

(1317) cnan remkel§-in poyg-cn m c c-tku-nin

he.ERG guest-POSS spear-ABS break-ITER-3SG/3SG.AOR
‘He broke the guest’s spear’ (neutral)
(1318) cnan poyg c-m a-tko-nen remk cl§-cn
he.ERG spear-break-ITER-3SG/3SG.AOR guest-ABS

126
It is doubtful whether we can speak of a primary target in (1303)-(1306).
 441

‘He broke the guest’s spear (into many pieces)’ (direct affect on the guest) (Chukchi, Polinsky
et al 1988: 683)

(1319) no-peku te tolida-§u

3R-backfist CORE cousin-2SG.POSS
‘He backfisted your cousin’
(1320) no-peku-§e na tolida-§u
3R-backfist-3OBJ NOM cousin-2SG.POSS
‘He backfisted your cousin’
(1321) no-peku-ko na tolida-§u
3R-backfist-2SG.OBJ NOM cousin-2SG.POSS
‘He backfisted your cousin’ (Tukang Besi, Donohue 1999bc: 377)

Both in Chukchi and Tukang Besi, the degree of affectedness associated with
possessors is sensitive to transitivity. In Chukchi, the direct affectedness is
associated with the absolutively marked Patient. In (1317), which illustrates the
typical transitive pattern, ‘spear’ is the primary Patient and is consequently viewed
as the directly affected participant. Whether the event has any effect on the ‘guest’
is irrelevant. In (1318), the possessor has been promoted in status and ‘spear’ is
incorporated into the verb. Despite its decreased status, ‘spear’ is still directly
affected by the event (perhaps even more so, as suggested by the alternations).
Instead of one affected participant, (1318) has two. Examples from Tukang Besi are
very similar to (1317)-(1318). Also in this case, an indirectly affected participant is
promoted in status. (1320) and (1321) differ from each other in that in (1321) ‘you’
is more directly affected than in (1320). Functionally, (1317)-(1321) are rather
similar to malefactive alternations illustrated above. We have, however,
distinguished between these two in the present context because of their structural
manifestation. In (1318) and (1321), nothing per se is added to clauses, but an
implicit participant is promoted in status. In Chukchi, the result is a typical transitive
construction (with an incorporated noun), whereas in Tukang Besi the number of
arguments increases, since ‘you’ is referred to by a person affix in (1321).
Furthermore, the alternations in (1317)-(1321) do not follow from a mere
introduction of an indirectly affected participant, but the conditioning factor is the
degree of affectedness associated with the participant.
So far, we have been illustrating alternations in which an increase in the
442

affectedness parameter causes a transitivity alternation. The last type to be discussed

here comprises cases in which transitivization is due to increased agency (cf. also
(156)-(157) from Marwari, (505)-(506) from Navajo and (876)-(877) from
Chepang), cf.

(1322) im-u- ’u-ru di=b arsi b=ita

father-OBL-CONTESS-EL I.OBL=GEN.N money N=be.lost.PAST
‘My father lost my money (guiltlessly)’
(1323) im-u-di di=b arsi b=it-ali
father-OBL-ERG I.OBL=GEN.N money N=be.lost-CAUS.PAST
‘My father lost my money (blamefully)’ (Godoberi, Kibrik 1996: 118)

(1324) lam cya-šc kataav c æhen cva

child-DAT story hear.PRES.PASS
‘The child hears the story’
(1325) lam cya kataav c ahan cva
child story hear.PRES.ACT
‘The child is listening to a story’ (Sinhala, Bynon et al 1995: 126f)

(1326) rai (INTR) ‘see’

rai-ca (TRANS) ‘see it’
va’a-rai-ca (CAUS) ‘watch, inspect’ (Boumaa Fijian, Dixon 1988: 189)

Events profiled in (1322) and (1324) appear typically in less transitive frames in
Godoberi and Sinhala, which correlates with less transitive semantics. The structural
transitivization attested in (1323) and (1325) follows from an agentivization. The
agent is directly responsible for having lost the money in (1323), whereas in (1325)
the child is actively involved in listening to a story. Examples from Fijian are
similar to Godoberi (cf. also (860)-(861)), since causativization of the verb can also
signal higher agency without introducing a participant to the given event.

5.3.2. Indirect alternations

So far, we have been illustrating alternations in which changes in semantic

transitivity are directly reflected in the morphosyntax of clauses. These are here
labelled as direct alternations. As opposed to these, alternations illustrated in the
 443

present section are reflected indirectly. This means that case marking and verb
morphology are not affected. The rationale behind these two types is the same, but
the alternations are only encoded differently. Both alternations can, for example, be
due to changes in the agency parameter. There are three different types of alternation
that will be labelled as indirect. Before proceeding, it is in order to say that most of
the examples are from myself or from informants familiar with the languages
exemplified. This is due to the fact that indirect alternations (as the term is
employed here) are often ignored in grammars.
The notion of indirect alternations comprises changes in some relevant
transitivity feature that are not (necessarily) reflected directly. Typical examples
include restrictions on passivization attested in many languages. As generally
known, passivization is in many languages restricted to clauses denoting rather
transitive events (see also 2.2.). For example, in English passives like *five apples
are contained by the jar are ungrammatical (for a detailed discussion of the topic in
light of Finnish and German, see Luukkainen 1988: 46-78, and for English see Rice
1987). It is, however, very important to note in the present context that these
restrictions must not be in any way conditioned by structural criteria. Consequently,
the ungrammaticality of deriving passives from intransitive clauses in English does
not qualify as an indirect alternation, since the ungrammaticality follows from the
structural intransitivity of the non-derived clause (cf. also (402)-(409) from Finnish
and German). Furthermore, the structural intransitivity is motivated by the semantics
of the event at issue, i.e. differences in semantic transitivity are marked also directly.
The latter violates the most important criterion for indirectness; the basic structure
of clauses has to be the same in order that we can speak of indirect alternations.
The first indirect alternation type is illustrated by cases in which two (or
perhaps even more) case frames are possible depending on differences in
transitivity. These are here labelled as indirect, since the alternation is not
necessarily reflected in any way, but changes in case frames are optional. It is also
of the utmost importance that an event that allows a certain kind of variation can
also appear in the typical transitive frame of a given language. If this criterion is not
met, we are dealing with a typical direct alternation. A couple of examples are
examined below, cf.
444

(1327) yiri-goyi-§molk gun-go §je gungu-dar §

1EXCL-know-PNEG gu-that gu-tree
‘W e didn’t know that tree’
(1328) yiri-mar õgi gun-go §je-§gen gungu-la õga-§gan
1EXCL-not know gu-that-DAT gu-billabong-DAT
‘W e don’t (didn’t) know that billabong’ (Ngalakan, Merlan 1983: 51)

(1329) m § *sudh ~/sudh ~l~ p ~hto

I Sudha-ACC see
‘I see Sudha’
(1330) mal ~ sudh ~/*sudh ~l~ diste
I-DAT Sudha/Sudha-DAT see
‘I see Sudha’ (Marathi, Pandharipande 1990: 169)

(1331) sa ye-wéhx ©
bear 3OBJ-killed
‘The bear killed him/her’
(1332) koR túé’ ye-wéhx ©
liquor 3OBJ-killed
‘The liquor killed him/her’
(1333) koR túé’ k’é lán ©we
liquor k’é died
‘S/he died from the liquor’
(1334) *sa k’é lán ©we
(S/he died from the bear) (Hare, DeLancey 1984b: 188)

(1335) mies viha-si nais-ta

man.NOM hate-3SG.PAST woman-PART
‘The man hated the woman’

(1336) mies jo-i maito-a / maido-n

man.NOM drink.PAST-3SG milk-PART / milk-ACC
‘The man was drinking the milk/drank the milk’
(1337) *mies viha-si naise-n
man.NOM hate-3SG.PAST woman-ACC
(The man hated the woman) (Finnish)

(1338) nøngmaq (nø) rømnvng ànglí tiq-cé-gø vdá-ì-‘

1PL TOP friend old one-ten-CL have-1PL-N.PAST
 445

‘W e have ten old friends’

(1339) àng pé zvt-‘
3SG basket weave-N.PAST
‘He weaves baskets’
(1340) àng-í pé tiq-chvng zvt-ò-‘
3SG-AGT basket one-CL weave-3.TR.N.PAST-N.PAST
‘He is weaving the basket’ (Dulong/Rawang, LaPolla 2000: 285ff)

Examples above have in common that semantic transitivity conditions the number
of syntactic frames in which a certain verb can appear. The examples diverge in
whether events higher or lower in semantic transitivity allow multiple frames. Also
certain other differences will be taken account of in the following discussion.
Examples from Ngalakan, Marathi127 and Hare (cf. also (1060)-(1067)) illustrate
cases in which events lower in transitivity allow multiple case frames. It is important
for our purposes that the alternation between frames in the less transitive cases is not
conditioned by relevant changes in transitivity. This means, for example, that (1329)
and (1330) can be used to describe the event in a rather free variation. What entitles
us to label this as an indirect alternation is that the possible variation is conditioned
by transitivity. It is restricted to semantically less transitive events. For example, as
explicitly shown in (1331)-(1334), the structure of highly transitive clauses, as
(1331) is identical to (1332) that displays lower transitivity. Only (1332) is eligible
for the less transitive frame illustrated in (1333). Examples from Finnish and
Dulong/Rawang differ in two major respects from (1328)-(1334). First, the variation
is determined by high transitivity. Furthermore, and foremostly, changes in case
frames of typically transitive events are conditioned by changes in semantic
transitivity. For example, in (1336) accusative and partitive are not in a free
variation, but the case marking of the Patient is conditioned by whether the event
is fully completed or not. What makes it justified to label examples like (1335)-
(1340) as similar to (1327)-(1334) is that two identical case frames behave
differently as regards whether they can appear in other (in this case highly transitive)
case frames. Hence, we may claim that we are dealing with an indirect alternation

127
W e have no evidence to explicitly show that the variation illustrated in (1329) and (1330) is
restricted to less transitive clauses, but this seems plausible.
446

in which higher semantic transitivity enables a verb to appear in two distinct case
frames. Both illustrated types underline the close relation of basic transitive clauses
and high semantic transitivity. In (1328)-(1334), less transitive constructions are
possible only, if the event denoted is low in transitivity. In (1335)-(1340), on the
other hand, a less transitive construction is obligatory, if the semantic transitivity is
low. Originally transitive events can also appear in this frame provided that the
transitivity is decreased. The opposite is not possible.
In (1327)-(1340), the (more or less) free variation follows from obvious
differences in semantic transitivity. For instance, in examples from Hare the degree
of agency associated with the instigator is radically lower, which makes it
impossible for (1331) to appear in a less transitive frame. Moreover, there are also
cases in which the conditioning factor is less directly related to obvious changes in
semantic transitivity, even if transitivity is the determining factor here as well.
Examples have been illustrated in (414)-(430) and (1135)-(1136). In these cases, the
nature of applicable alternations is not sensitive to obvious changes in semantic
transitivity, but the relevant factor is the motivation of transitivity. In Finnish, Malay
(cf. also below), Tukang Besi and Inuktitut inherently transitive and causative
behave differently as regards the derivation of IAC’s or antipassives. The semantic
transitivity of events per se is similar, but is motivated differently. Since these kinds
of difference are only indirectly related to differences in semantic transitivity, they
are not studied in any more detail in the present context, but they still merit a note.
The type referred to above leads us to discussing the second type of indirect
alternation that comprises cases in which two identical constructions diverge in the
kind of transitivity alternations applicable to them. This is also conditioned by
semantic transitivity. Hence, examples (1335)-(1340) share some common features
with this type, since the variation illustrated in (1336) and (1339)-(1340) is
conditioned by semantics as well. The inability of semantically less transitive
clauses to passivize is an illustrative example of this alternation type. A couple of
further examples are illustrated below, cf.

(1341) yo balyé lakou-la

they sweep yard-DEF
‘They swept the yard’
 447

(1342) lakou-la balyé

yard-DEF sweep
‘The yard was swept’
(1343) *on dòt mèb ka fèt
a other piece of furniture NONPAST made
‘Another piece of furniture is being made’ (Lesser French Antillean Creole, Gadelii 1997:
109)

(1344) sa-jumúútya(-ráy)
3SG-help(-1SGO)
‘She is helping (me)’
(1345) sa-rícha-rà
3SG-extinguish-INANO
‘S/he extinguished it’
(1346) ?sa-rícha
‘S/he extinguished’ (Yagua, Payne 1985: 33ff)

(1347) der mann /das baby /die katze ha-t

ART.NOM man /ART baby /ART cat have-3SG.PRES
den teller zerbrochen
ART.ACC plate break.PARTIC
‘The cat/the baby/the man broke the plate’
(1348) dem mann /?dem baby /??der katze ist
ART.DAT man /ART.DAT baby /ART.DAT cat be.3SG.PRES
der teller zerbrochen
ART.NOM plate break.PARTIC
‘The man/?the baby/??the cat broke the plate accidentally’ (German)

(1349) orang /bayi itu pecah-kan pasu itu

person /baby that break-CAUS vase that
‘That person/baby broke the vase’
(1350) orang /?bayi itu ter-pecah pasu itu
person /baby that ter-break vase that
‘That person/?baby broke the vase accidentally’ (Malay, examples courtesy of Foong Ha Yap)

(1351) carlos jacinto shpo-no

Carlos Jacinto hit-PERF
‘Carlos hit Jacinto’
448

(1352) jacinto shpo-ro-r

Jacinto hit-PERF-1SG
‘I hit Jacinto’
(1353) domer e op zrö-no e
man DEM REFL kill-PERF CFP
‘The man, he killed himself’
(1354) *op zrö-r-a e
REFL kill-PERF-3SG CFP
‘(He killed himself)’ (Teribe, Quesada 2000: 107, 144)

(1355) ich schlag-e dich / mich

1SG.NOM hit.PRES-3SG 2SG.ACC / 1SG.ACC
‘I hit you/myself’
(1356) ich befürcht-e mich
1SG.NOM be.afraid.PRES-3SG 1SG.ACC
‘I am afraid’
(1357) dich / mich schlag-e ich
2SG.ACC / 1SG.ACC hit.PRES-3SG 1SG.NOM
‘I hit YOU/MYSELF’
(1358) *mich befürchte ich

Examples from LFAC illustrate a further example of ungrammatical passives. In

LFAC the ability of clauses to derive passives is conditioned by the aspect (defined
semantically) of clauses. Only clauses in the perfective aspect can be passivized,
hence (1343) is ungrammatical. Another factors of this kind include lower inherent
transitivity of clauses and the nature of agent. For example, in Finnish the
(obligatorily unexpressed) agent has to be human or higher animate in order that
passivization is possible. In Kayardild (see Evans 1995: 339), ‘objects’ of emotion
verbs cannot be promoted to subjects of passives (this is typically also the case in
languages like English and German). Examples from Yagua illustrate a case in
which the Patient omission is possible only if the event denoted is not typically
transitive.128 It is an interesting detail that we have not come across similar
restrictions on antipassivization. This might be simply be due to the fact that this
aspect of antipassives is not discussed. Kibrik (1996: 139f) is the only detailed

128
This is a generalization with some exceptions, as shown by Payne.
 449

discussion of this aspect of antipassivization (in Godoberi). However, based on the

findings of Kibrik, it does not seem that high transitivity per se would be the
conditioning factor, since certain intransitive verbs allow antipassivization as well.
Consequently, an independent study is sorely needed to shed more light on this
topic. Examples from German and Malay exemplify cases in which obvious changes
in agency are reflected in whether an IAC alternation can be applied to clauses. Only
if the agent is highly agentive to begin with can an IAC be applied to a clause. Cats
and babies can occupy the Agent slot in typical transitive clauses, as shown in
(1347) and (1349). Their lower agency is reflected only indirectly. IAC’s are
infelicitous or less acceptable, if a transitive event is instigated by a baby or a cat.
In German and Malay (perhaps along with other languages), IAC-alternations imply
that the agent is highly agentive. This follows, since the function of this alternation
is to decrease agency. Hence, if the agent is lower in agency, IAC’s are not possible.
In Teribe, Agents of typical transitive clauses ((1351)) are signalled only on the
verb. Reflexives are syntactically transitive constructions in the language, since
reflexivity is expressed by replacing the transitive Patient by a general reflexive
particle. The lower transitivity of the clause is reflected only indirectly. A somewhat
similar case is attested in German in which true (semantic) reflexives129 and middle
clauses differ from each other in the status of the Patient. The grammatical status of
the Patient is obviously higher in reflexives than in middle clauses (see Grewendorf
1988: 54ff and Bausewein 1990: 68 for a more detailed analysis), even if it is not
as high as that of typical Patients (reflexive pronouns cannot, for example, be
promoted to subjects of passives). Morphologically, both are identical to typical
Patients, which makes also middle clauses similar to typical transitive clauses in this
respect.
Examples (1341)-(1358) illustrate cases in which lower semantic transitivity
results in ungrammaticality of certain operations related to the expression of
transitivity. For example, passivization is in many languages restricted to clauses
that are sufficiently transitive. The affectedness of the patient and the agency of the

129
Traditionally ,these are labelled as ‘false reflexives’ (unechte Reflexivverben, cf. Helbig und
Buscha 1993: 65), while middle clauses as ‘true reflexives’ (echte or inhärent reflexive Verben, cf.
Helbig und Buscha:i bidem and Grewendorf 1988: 64f).
450

instigator are usually relevant in this respect. The opposite of this is also possible.
This refers to cases in which events inherently rather low in transitivity are
implicitly transitivized. For example, clauses that typically exclude passivization
can be manipulated in order to make passivization possible. A typical example of
this is provided by passives like you were seen and the corresponding structures in
other languages. In some languages (including English, Finnish and German),
passives like these are typically used only if the stimulus is regarded as patient, and
is affected by the event. This means that ‘being seen’ adversely affects the entity
referred to. This kind of passive is appropriate, for example, in case a thief has been
seen by someone when committing a crime, which may result in an arrest. These
kinds of indirect alternation are rather problematic, since the context is always
manipulated for making the given construction grammatical. We are dealing with
two different interpretations of one basic event, i.e. the inherent transitivity of events
does not aid us in distinguishing grammatical cases from ungrammatical ones,
which is possible in (1341)-(1358). In order to regard these as genuine indirect
alternations, we would have to be able to show that the affectedness of the patient
really is higher and that the passivization is possible only if this condition is met.
Since the passiviability of semantically more transitive clauses is the only applicable
criterion here and (only) it manifests itself, circularity cannot be avoided.
Alternations illustrated in (1341)-(1358) provide us with direct evidence for
the fact that the function of transitivity alternations like passive and antipassive is
to decrease the transitivity of clauses. These alternations are in many languages
excluded in case the overall transitivity of clauses is low to begin with. This is very
natural, since if the function is to decrease transitivity, the transitivity of the original
clause has to be rather high. The third kind of indirect alternation is represented by
cases in which alternations like passive and antipassive are possible, but the derived
structures are sensitive to transitivity of the underlying clauses. According to
Drossard (1991: 447), certain differences are made semantically explicit
(semantisiert) as a result of passive or antipassive derivation. Since we are dealing,
e.g. with typical passives and antipassives in these cases, alternations illustrated
above could also be labelled as direct. There is a basic construction with a derived
less transitive counterpart. However, in cases presented below, only derived
constructions reflect the relevant semantic differences. The conditioning factor of
 451

the alternations is very similar to direct alternations. The primary difference between
these types lies in how the differences are expressed. In the third type of indirect
alternations, the basic (i.e. non-derived) structure is neutral for transitivity, while the
derived constructions are sensitive to it.130 Examples follow, cf.

(1359) tinoni pia va-opo-na ia mola

man this CAUS-capsize-3SG.OBJ ART.SG canoe
‘This man capsized the canoe’
(1360) ta-va-opo teia tinoni pia ia mola
PASS-CAUS-capsize AGT man this ART.SG canoe
‘The canoe was capsized by this man’
(1361) bolusu va-opo-d i ria mola
big.wave CAUS-capsize-3PL.OBJ ART.PL canoe
‘A/the big wave capsized the canoe’
(1362) ta-va-opo bolusu ria mola
PASS-CAUS-capsize big.wave ART.PL canoe
‘The canoes were capsized by a/the nig wave’
(1363) ta-va-lequ pa kalakea tinoni ia bue pia
PASS-CAUS-be.dead AGT some man/person ART pig this
‘This pig was killed by somebody’ (Vangunu, examples courtesy of Frank Lichtenberk)

(1364) ta-nani-ni-a rao sa leboto

PASS-bite-APPL-3SG 1SG ART machete
‘Her leg was cut by the machete’
(1365) ta-hakeagi te amina rao
PASS-look.after PREP Amina 1SG
‘I am looked after by Amina’
(1366) sa baha sagi koni-ta-baha-ni-a goe
SG.ART witch TOP FUT-PASS-bewitch-APPL-3SG 2SG
‘The witch, you’ll be bewitched by her’ (Hoava, Davis 1997: 261ff)

(1367) der hund hat mich gebissen

ART.NOM dog have.PRES.3SG 1SG.ACC bite.PARTIC
‘The dog bit me’
(1368) ich wurde von (*durch/mit) dem hund
1SG.NOM become.PAST.3SG ‘by’ ART.DAT dog

130
The same topic is discussed by Drossard (see 1991: 467ff) in light of German and Yidiñ.
452

gebissen
bite.PARTIC
‘I was bitten by the dog’
(1369) der blitz steckte das haus in Brand
ART.NOM lightning set.PAST.3SG ART.ACC house in fire
‘The lightning set the house on fire’
(1370) das haus ist von (*mit) dem /durch
ART.NOM house be.PRES.3SG by ART.DAT /through
den blitz in brand gesteckt worden
ART.ACC lightning in fire set.PARTIC become.PARTIC
‘The house was set on fire by the lightning’
(1371) bombe-n zerstörte-n die stadt
bomb-PL.NOM destroy.PAST-3PL ART.ACC city
‘Bombs destroyed the city’
(1372) die stadt wurde von /durch /mit bomben
ART.NOM city become.PAST.3SG ‘by’ /through /with bombs.PL
zerstört
destroy.PARTIC
‘The city was destroyed by bombs’ (German, with slight changes from Drossard 1991:467)

(1373) no-nabu te kaluku na amai ito

3R-drop CORE coconut NOM 3PL that.higher
‘They dropped the coconut’
(1374) no-to-nabu-mo na kaluku
3R-PASS-drop-PFV NOM coconut
‘The coconut was dropped (by someone)’
(1375) no-te-nabu-mo na kaluku
3R-ACCID.PASS-drop-PFV NOM coconut
‘The coconut happened to fall’ (Tukang Besi, Donohue 1999: 279)

(1376) joo » na-yíí-»-ne §

ball PREF-3.ACC/3.NOM .MD-TI-SRO.move.PF
‘He dropped the ball’

(1377) joo » naa-l-ne §

ball PREF.3NOM .MD-TI-SRO.move.PF
‘The ball was dropped by someone’
(1378) *joo » na-bi-§-doo-l-ne §
ball PREF-3.ACC-IND.NOM -TD.MD-SRO.move.PF
 453

(The ball was dropped)

(1379) §asd '' §ashkii tá-né-í-Ø-z-Ø-giz
woman boy PREF-PREF-3.ACC-3.NOM-PFV-TI-wring.PFV
‘The woman washed the boy’
(1380) §ashkii táá-bí-§-dí-s-d-giz
boy PREF-3.ACC-IND.NOM -TD-MD-TI-wring.PF
‘The boy was washed’ (Navajo, Kibrik 1996b: 266, 275f, 287)

(1381) wagud ya-ngu bun ya wawa-l

man-ERG woman look at-PRES
‘The man is looking at the woman’
(1382) wagu:d ya bun ya:-nda wawa-:d yi-õ
man woman-DAT look at-ANTIP-PRES
‘The man is looking at the woman’ (Yidiñ, Dixon & Blake 1979b:14)
(1383) waguja-ngu jugi-ø gunda-l (galba:n-da)
man-ERG tree-ABS cut-PRES axe-INSTR
‘The man is cutting a tree (with an axe)’
(1384) wagu:ja-ø gunda-:ji-õ (jugi-:l) galba:n-da
man-ABS cut-ANTIP-PRES (tree-LOC) axe-INSTR
‘The man is cutting a tree with an axe’ (Yidiñ, Dixon 1994: 59f)

(1385) bama-lu gurra: du:-ny

man-ERG dog.O hit-PAST
‘The man hit the dog’
(1386) bama gurra:-nda du:-yi-ny
man dog-DAT hit-INTR-PAST
‘The man hit the dog’
(1387) yaba-nggu djulbin guni-l
brother-ERG tree.O cut-PRES
‘Brother cuts a tree’
(1388) yaba djulbin-da guni-yi-ng
brother tree-LOC cut-INTR-PRES
‘Brother cuts a tree’ (Djabugay, Patz 1991:299)

Examples above illustrate passive and antipassive derivations that reflect differences
in semantic transitivity. The features are irrelevant in basic clauses. Examples
(1359)-(1380) present differences in passivization, whereas (1381)-(1388) illustrate
similar antipassives. In Vangunu, Hoava and German the verb morphology of
454

passives is constant. The different instances of passive are distinguished on the basis
of Agent marking. In all three languages, animacy of the agent is very important.
In Vangunu, the definiteness of the agent also seems to make a contribution (cf.
(1360) and (1363). In Hoava, the expression of (definite) inanimate Agents requires
that the verb be applicativized.131 Animate Agents do not require applicativization,
but the Agent is preceded by the preposition te, as in (1363). (1364) illustrates the
marking of Agents ‘regarded as something less than a real human’ (Davis 1997:
263). As in the case of definite inanimate Agents, applicativization is possible in
this case. Vangunu and Hoava exemplify languages in which semantic differences
are necessarily reflected through distinct marking of Agents in passives. The nature
of passivization is different in these cases. German, on the other hand, illustrates a
language in which the number of applicable mechanisms underlines the semantic
differences. If the Agent refers to a typical agent, as in (1367) and (1368), only von
can mark the Agent. In case the instigator is a ‘force’, durch along with von is
possible, too. Instruments that can be labelled as the least typical agents allow the
Agent to be marked in three different ways, as shown in (1372). Examples from
German illustrate an indirect version of (1060)-(1067), since the higher number of
possible Agent marking mechanisms correlates with a lower transitivity of events.
Examples from Tukang Besi exemplify a distinction of different passives based on
the verb morphology alone. According to Donohue (ibidem), the difference attested
in (1374) and (1375) is due to animacy. The passive is derived by the affix to-, if the
(in this case unexpressed) agent is a typical human agent. (1375) is possible only if
the verb can take a generic or ‘natural’ actor, i.e. they do not require agency or
volition (Donohue 1999: 278). The appropriate translation is ‘the coconut happened
to fall’, as suggested by Donohue. In Navajo, there are two distinct passives labelled
as ‘passive’ and ‘semi-passive’ by Kibrik (Kibrik 1996b: 266, 275). These differ
from each other as regards the nature of the patient involved. In the case of
inanimate undergoers, (typical) passive is the only possible construction. Semi-
passive is ungrammatical, as shown in (1378). Semi-passive requires that the
undergoer be animate. Hence, (1380) is deemed grammatical. In case the undergoer
is human, only semi-passive is possible, while animate non-human undergoers (i.e.

131
Indefinite Agents of this kind can be introduced to clauses without applicativization.
 455

animals) allow both passive types (see e.g. Kibrik 1996b: 287f). The typical passive
can be considered more transitive semantically, since the patient is a passive
participant that cannot control the event in any way. This is also reflected
structurally, since the accusative marking has been retained in the semi-passive,132
which makes the semi-passive less-of-a-passive also structurally. Examples from
Yidiñ and Djabugay present antipassives that differ from each other based on
animacy (and inherent definiteness) of the Patient. Animate Patients of appear in the
dative in antipassives, while inanimate Patients are marked in the locative. As in
(1361)-(1380), this distinction is neutralized in basic clauses.133 In Yidiñ, the dative
marking is obligatory, if the Patient is human and thus inherently definite. In the
case of non-human animates, both dative and locative are equally possible (it is not
discussed by Dixon, whether this variation is conditioned by definiteness or not),
whereas locative marking is clearly preferred for inanimate Patients in antipassives
(Dixon 1977: 277). The marking of Patients is identical in Djabugay. It is perhaps
noteworthy in this context that according to Patz (ibidem), there are no semantic
differences between typical transitive clauses and antipassives in cases illustrated
above. Hence, we are in a way dealing with two different ways of expressing a
single event (cf. (1327)-(1340) above). Semantic transitivity (i.e. definiteness of
Patients) is relevant only in derived, syntactically less transitive clauses.

5.3.3. Structurally determined alternations

In the two previous sections, we have illustrated alternations due to obvious changes
in semantic transitivity. In what follows, we will briefly illustrate a further
transitivity alternation type that differs from the others under study in that their use
is not determined by any facet of semantic transitivity, but they are primarily due to
grammatical requirements of a given language. These kinds of alternation are
expressed in principle by the same mechanisms as direct alternations illustrated

132
On the other hand, the patient is definite in semi-passive, which enables us to argue for its higher
transitivity.

133
-lu and -nggu are different realizations of the same ergative function, this variation is not
conditioned by semantics.
456

above, but the motivation is different. Hence, as structural are here labelled
alternations that are structurally similar to direct alternations (and also indirect
alternations illustrated last), but which are conditioned by radically different factors.
There are three types of structurally conditioned alternations that are all illustrated
and discussed below. As is the usual practice in the present study, we begin by
elaborating the brief definition given here.
As noted above, structurally conditioned alternations are primarily required
by grammaticality of clauses. This makes them an integral part of the grammar of
a given language. They are applied for the purpose of avoiding ungrammaticality.
It is of the utmost importance to note that to be labelled as primarily structural, an
alternation must not be conditioned by any aspect of semantic or structural
transitivity of clauses. This note is in order to exclude from the discussion here cases
like (280)-(285) in which the syntactic transitivity gives rise to the transitive verb
morphology. In these cases, transitivity of the verb is conditioned by the number of
explicitly expressed arguments which can be considered as the most fundamental
facet of structural transitivity. Hence, a certain feature of structural transitivity
reflects another one and is conditioned by it. Moreover, both mechanisms are
directly related to transitivity and express transitivity explicitly. In the cases
presented in the present section, on the other hand, a certain feature of transitivity
relies on another one without being directly inferable from it. The given features are
not directly related to transitivity in any way, but they are rather submitted to it.
The first type of structural alternation is presented by cases in which
(structural) transitivity of clauses is conditioned by the clause type. The marking in
main clauses is usually more transitive than that in subordinate clauses which is
illustrated below, cf.

(1389) ngurra ngatha nhanya-nyja thana-mpa-nha

camp.ACC 1SG.ERG see-PAST 3PL-DAT-ACC
‘I saw their camp’ (Jiwarli, Austin 1997b:26)
(1390) papu ngathi-nyja mura-kura-wu parnu-mpa-wu
father.NOM cry-PAST son-own-DAT he-DAT-DAT
‘The father cried for his son’ (Jiwarli, Austin 1997b: 8)
(1391) kaji nhurra yana-ma mana-ngku ngurlu karla-rla
try 2SG.NOM go-IMPER get-PURP.SS that.ALL fire-ALL
 457

‘You try and go to get the fire’

(1392) ngatha papa-ngku-rru puntha-rninyja pirtura-rni-rnu karla-wu
1SG.ERG water-ERG-now douse-PAST extinguish-CAUS-IMPF.SS fire-DAT
‘I doused (him) with water and put the fire out’ (Jiwarli, examples courtesy of Peter Austin)

(1393) thuthu-ngku jarti-lanyi mantu

dog-ERG eat-PRES meat
‘The dog is eating the meat’
(1394) ngatha wayanpi kartanhawu nguurru-wu
1SG.NOM afraid that.DAT horse-DAT
‘I am afraid of that horse’
(1395) ngatha-rna nyina-nyi papa-wu walha wula-rnuru
1SG.NOM -1SG.S sit-PRES water-DAT firs t drink-REL SS
‘I am sitting drinking water, to begin with’ (Yingkarta, Dench 1998: 19, 22, 70)

(836) tuka-yu tuar ityayi

dog-ERG snake bite
‘The dog bites/bit the snake’
(837) tuku (tuar-ku) ityayi
dog.ABS (snake-DAT) bite
‘The dog is biting (the snake)’ (Kalkatungu, Blake 1982: 86) (repeated here for convenience)
(1396) nanya nga-thu kalpin thuku-ku lha-yi-nyin
saw I-ERG man dog-DAT hit-ANTIP-PART
‘I saw the man hitting the dog’ (Kalkatungu, Blake 1987: 148)

In the languages cited above, the argument marking in subordinate clauses is less
transitive than that in main clauses. In Jiwarli, the Patient appears in the accusative
(or the absolutive) in main clauses, while the marking in subordinate clauses is
dative or allative. In Yingkarta and Kalkatungu, the marking of the Patient shifts
from the absolutive to the dative in subordinate clauses. In all of these languages,
the dative marking of the Patient is also used to signal lower degrees of transitivity,
as shown in (1390), (1394) and (837). What distinguishes (1391), (1392), (1395)
and (1396) from these examples is the underlying motivation of the less transitive
Patient marking. In the latter cases, the shift from accusative or absolutive to dative
or allative does not reflect any changes in semantic transitivity, but it is purely
structurally motivated. The employed mechanism per se is the same in both cases,
because of which these cases are taken account of in the present context.
458

In Jiwarli, Yingkarta and Kalkatungu, the marking of subordinate clauses is

less transitive than that of transitive main clauses. This is very likely the norm in
languages displaying a split conditioned by clause type. For example,
nominalizations or different kinds of converb are typically attested in subordinate
clauses, which is often accompanied by a de-transitivization of some kind (the range
is from mildly affected cases to complete nominalizations of subordinate clauses in
which case the marking of both arguments along with verb morphology is affected).
Cavineña, on the other hand, illustrates a counter-example, since the shift is from
lower to higher structural transitivity in subordinate clauses, cf.

(1397) tu ja ya-tse waka k wana ba-tsa-t •ine umada

so 1-DL.ABS cow PL.ABS see-arriving.O-PAST many
‘So we two saw many cows coming’
(1398) i-ke ni mi-ra ara-kara a-ya duhu-k we i-ke
1-AFFIX.ABS if 2-ERG eat-want do-PRES take-IMPER 1-AFFIX.ABS
espere hiruru
stream edge
‘If you want to eat me, take me to the edge of the stream’ (Cavineña, Camp 1985: 42ff)

According to Camp (1985: 44), the ergative marking of Agent is obligatory for
nouns, whereas pronouns appear in the ergative optionally. (1397) illustrates a case
in which a pronoun does not appear in the ergative. What makes Cavineña
interesting is the shift from optional to obligatory ergative marking of Agent
pronouns in subordinate clauses. In (1398), the ergative marking on mi-ra is not
omissible without this resulting in ungrammaticality. Thus, subordinate clauses with
an Agent pronoun outrank corresponding main clauses in transitivity. In the latter,
the ergative marking is optional. Subordinate clauses are obligatorily transitive (as
regards the marking of Agent), while high transitivity is optional (yet rather frequent
in light of data cited) in main clauses.
In (1389)-(1398), alternations are conditioned by the clause type arguments
appear in. Somewhat different examples of pure structurally conditioned alternation
are illustrated below, cf.
 459

(1399) nau ku qani-a botho...

1SG 1SG.NFUT eat(TRANS)-3SG.OBJ pork...
‘I ate pork...’
(1400) nau ku fanga baqita qana alo
1SG 1SG.NFUT eat(INTR) be.big PREP taro
‘I ate a lot of taro’ (Toqabaqita, examples courtesy of Frank Lichtenberk)

(1401) inau na-lahi pilu-ni puuke ona-ku

1SG 1SG.REAL-carry stick.together-TRANS book POSS-1SG
‘I carried my books in one hand’
(1402) *inau na-lahi-nV puuke ona-ku pilu
1SG 1SG.REAL-carry-COMM.OBJ book POSS-1SG stick.together
(I carried my books in one hand) (Paamese, Crowley 1987: 63)

(1403) à:ng-í shøng rí-mvn-ò-‘

3SG-AGT tree/wood carry-continue-3.TR.N.PAST-N.PAST
‘He is continuing to carry the wood’
(1404) àng yøp-mvn-shì-‘
3SG sleep-continue-REFL/MID-N.PAST
‘He is continuing to sleep’ (Dulong/Rawang, LaPolla 2000: 293f)

In distinction from (1387)-(1398), the relevant feature of transitivity above is the

verb morphology. In all these languages, (in)transitivity of certain verbs is a
structural necessity. In Toqabaqita and Paamese, not every combination of transitive
and intransitive verbs are possible in serial verb constructions, which produces
structurally motivated transitivity alternations. In Toqabaqita, the first verb of every
serial verb construction is obligatorily intransitive. Consequently, in (1400) the verb
‘eat’ appears in a suppletive intransitive form without an object marker -a. In
Paamese, on the other hand, every verb of a ‘nuclear layer serialization’ has to be
transitive, since the verbs of these constructions cannot independently mark their
objects (Crowley: ibidem). Hence, in (1401) the transitivizing affix -ni has to be
attached to the verb. The intransitivity or transitivity of verbs follows from the
position of verbs in the constructions in question. Hence, the shift from transitivity
to intransitivity is purely structurally motivated. In Dulong/Rawang, the reflexive
(i.e. intransitive) marking of the verb in (1404) is not related to reflexivity, but it is
needed to intransitivize the auxiliary verb that has to match the matrix verb in terms
460

of transitivity (LaPolla: ibidem).

In (1389)-(1404), we are dealing with genuine structurally motivated
transitivity alternations with no semantic function whatsoever. The second type
comprises cases in which the function of a given alternation is to enable the
application of certain grammatical operations, cf.

(1405) bayi ya a bani u ba õgun dS ugumbi u balgan

ART man.ABS come.TNS ART.ERG woman.ERG hit.TNS
‘The man came in and the woman hit him’
(1406) bayi yara bani u bagun dS ugumbilgu balgalõa u
ART man.ABS come.TNS ART.DAT woman.DAT hit.AP.PAST
‘The man came in and hit the woman’ (Dyirbal, Dixon 1972: 130)

(1407) ó v déékaa mekó aka é

man PFV.go jaguar kill DECL
‘The man went and [the man] killed the jaguar’
(1408) ó v déékaa mekó óv aka é
man PFV.go jaguar man kill DECL
‘The man went and the jaguar killed the man’ (Suruí, Rodrigues 1999:1 22, cited from Van
der Meer 1985: 210)

(1409) ny mpianatra izay nahita ny vehivahy

the student REL saw the woman
‘The student that saw the woman’
(1410) *ny vehivahy izay nahita ny mpianatra
the woman REL saw the student
(The woman that the student saw)
(1411) ny vehivahy izay nohitan’ ny mpianatra
the woman REL seen the student
‘The woman that was seen by the student’ (Malagasy, Keenan & Comrie 1977)

(1412) ekikopo john ky’-akuze

cup John REL-brought
‘The cup which John brought’
(1413) ekiso john kye-yattisa enkonko
knife John REL-killed.INSTR.APPL chicken
‘The knfie with which John killed the chicken’
 461

(1414) *ekiso john kye-atta enkonko na

knife John REL-killed chicken with
(The knife with which John killed the chicken) ‘(Luganda, Keenan & Comrie 1979: 341)

(1415) umugabo a-ra-andik-a íbárúwa n’ííkárámu

man he-PRES-write-ASP letter with.pen
‘The man is writing a letter with the pen’
(1416) *íkárámu i-ra-andik-w-a íbárúwa n’ûmugabo
pen it-PRES-write-PASS-ASP letter by.man
(The pen is used to write a letter by the man)
(1417) íkárámu i-ra-andik-iish-w-a íbárúwa n’ûmugabo
pen it-PRES-write-INSTR-PASS-ASP letter by.man
‘The pen is used to write a letter by the man’ (Kinyarwanda, Kimenyi 1980: 81)

Examples from Dyirbal and Suruí illustrate typical examples of structurally

motivated transitivity alternations the function of which is to facilitate clause
coordination. In Dyirbal, clauses like (1405) has to be antipassivized (illustrated in
(1406)) in order to make it possible for ‘woman’ to be the Patient. This kind of
alternation is attested also in ‘non-exotic’ languages like English and German in
which passive is used in a similar function to create pivots (e.g. he came in and was
hit instead of he came in and somebody hit him or er kam hier und wurde
geschlagen instead of er kam hier und jemand schlug ihn). Suruí differs from
Dyirbal, English and German in that the verb morphology is not affected in (1407)
and (1408). Similarly to English and German, the omission is restricted to Agents
in the latter clause. If the sole argument of the preceding clause is co-referential with
the Patient of the latter clause, it has to be explicitly expressed. The latter clause is
thus more transitive. Examples from Malagasy and Luganda illustrate what has
been labelled as ‘noun accessibility hierarchy’ by Keenan and Comrie in their 1977
and 1979 papers. In these two languages, only subjects and objects (or primary and
secondary terms) can be relativized. Consequently, other terms have to be promoted
to the primary term status to enable relativization. Examples from Kinyarwanda are
rather similar to those from Malagasy and Luganda. In Kinyarwanda, arguments in
instrument function have to be promoted to object status via applicativization in
order that the argument in question can be promoted to subject of passive.
Alternations illustrated above differ from those presented in (1389)-(1404) in that
462

the alternations are not obligatorily required by the structure of the given languages,
but the alternations are necessary only in certain cases. For example, in Malagasy
and Luganda relativization per se does not require passivization or applicativization,
but this is necessary only if the relativized argument is not a subject. In similar vein,
in Dyirbal antipassivization is required only if the intended interpretation of the
clause in that given in (1406). In languages like Dyirbal and Suruí, the alternations
have a clear semantic function, but since this is not related to transitivity, these are
labelled as structurally conditioned alternations in the present context.
The third type of transitivity alternation labelled as primarily structural here
is presented by alternations in which a certain alternation is necessary for another
to be applied. Examples include cases in which certain alternations (expressed by
manipulating the verb morphology) are a prerequisite for others to be applied. These
alternations are functionally similar to those presented in (1405)-(1417), since the
alternations are not obligatory per se, but they are needed only in case certain
alternations have to be expressed, cf.

(1418) ni c cn q’ wcl-cm-stcx w ›c s |eni § §c k w›c s cplíl

AUX 1SG bake-ANTIP-CAUS DET woman OBL DET bread
‘I had the woman bake the bread’
(1419) *ni c cn q’ wcl-ct-stcx w k w›c s cplíl §c |c s |eni §
AUX 1SG bake-TR-CAUS DET bread OBL DET woman
(I had the woman bake the bread) (Halkomelem Salish, Song 2001: 269, cited from Gerdts
1984: 195)

(1420) i-’u’u-kur-’am-ban
1SG.SUBJ.2SG.OBJ-baby-hold-CAUS-PAST
‘I made you hold the baby’
(1421) *’u’ude i-kur-’am-ban
baby 1SG.SUBJ.2SG.OBJ-hold-CAUS-PAST
(I made you hold the baby) (Southern Tiwa, Song 2001: 269, cited from Baker 1988: 194f)

(1422) e pasi-tak-a na polo

3SG pass-TRANS-3SG DEF ball
‘He passed the ball’ (Fijian, Schütz 1985: 142)

Examples from Halkomelem Salish and Southern Tiwa illustrate very typical
 463

examples of the last type under study here. In both cases, causativization is possible
only if the construction is intransitivized first. In Halkomelem, the verb has to be
antipassivized prior to causativization, while in Southern Tiwa, causativization is
possible only if the Patient of transitive clauses is incorporated into the verb.
Causativization of transitives comprises consequently two independent alternations,
both of which have independent functions in the given languages. In these particular
cases, one of them is required to make the application of the other grammatical. In
this respect, these cases differ, for example, from typical passives in which the
deletion of Agent is possible only if the verb morphology is manipulated. In the case
of passive, we are, however, dealing with a single transitivity alternation comprising
two parts. Furthermore, causativization as such does not require that the verb be
antipassivized in Halkomelem, or that the Patient be incorporated in Tiwa, but this
is restricted to causativization of transitives. If causativization in languages like
Halkomelem always necessitated antipassivization, we would not be dealing with
a genuine structurally motivated alternation, since causativization could be regarded
as a single alternation that consists of two different parts. Examples from Fijian are
different. In Fijian, the general transitivizing affix -taki has to be attached to every
loan word to enable their transitive use. The affix does not attach to original Fijian
words in the same function. The affix has other (semantic) functions as well, but in
(1422) the function is simply to make the transitive use of loan words possible.
Alternations in (1418)-(1421) are rather similar to (1405)-(1417). These are
here distinguished based on the motivation of the ‘secondary’ alternation. For
example, in Kinyarwanda, passivization of three-argument constructions does not
necessarily require applicativization. This is necessary only if we wish to promote
a non-core argument to the subject of passive. Furthermore, the applicativization
serves its primary function also in (1417) (i.e. it promotes an instrument to object
status) and it is applied independently of passivization. Passive alternations do not,
in any direct or necessary way, depend on applicativization, whereas causativization
of transitives relies on antipassivization in Halkomelem. It is also important to
distinguish alternations like (1418)-(1421) from the likes of (724)-(734). The latter
kinds of alternations also consist of two different mechanisms, but differently from
(1418)-(1421) one of the alternations is motivated semantically, which makes it
impossible to label them as purely structurally motivated alternations. In (1418)-
464

(1421), on the other hand, antipassivization or Patient incorporation is

grammatically required. Moreover, they are not associated with their primary
function in these cases.
As noted in section 5.1., (exx. (352)-(356) repeated below for convenience),
languages in which nouns and pronouns follow different argument marking patterns
pose a possible problem for our definition of transitivity alternations. In certain
cases, differences in reference to participants in events cause changes that
structurally resemble transitivity alternations, cf.

(352) õura wanguli-la

dog bark-PAST
‘A dog barked’
(353) õura-õu munda bada-la
dog-ERG snake bite-PAST
‘A dog bit a snake’
(354) õaya barri-la
I cry-PAST
‘I cried’
(355) õaya na õu-na bada-la
I he-ACC bite-PAST
‘I bit him’
(356) õaya munda bada-la
I snake bite-PAST
‘I bit a snake’ (Bidjara, Blake 1976: 282, cited from Breen 1973)

In this particular case, (356) could be regarded as an intransitivizing alternation,

since in comparison with (353) and (355), either the explicit marking of Agent
((353)) or that of the Patient ((355)) is omitted. Or, in case (356) illustrates the basic
transitive pattern, (353) and (355) can be regarded as transitivizing alternations. If
we wish to include these kinds of case in our typology, the most appropriate way is
to analyze these examples as structurally motivated alternations, even if the
motivation is radically different from other illustrated cases. The semantics of events
is not affected in any way in (352)-(356), but the changes in the case marking follow
exclusively from the grammatical nature of arguments. The argument marking is
conditioned by the nature of reference to participants in events. Differently from
 465

passives and antipassives, the perspective per se is the same. Since the grammatical
nature of nouns and pronouns are responsible for ‘transitivity alternations’ like
(353), (355) and (356), these cases can be regarded as instances of structurally
motivated alternations. Differently from (1389)-(1422), the alternation is
conditioned ‘locally’, i.e. the nature of arguments is alone responsible for the
structural differences.

5.3.4. Final remarks

Above, we have proposed a semantic-functional typology of transitivity alternations.

The goal of the section has been to discuss the rationale behind transitivity
alternations as exhaustively as possible. Similarly to the previous typologies, we end
the presentation by illustrating the most important findings of the previous sections
schematically. The goal of this is to give the reader an easier access to the typology
without presenting anything new. The notation goes as follows. Labels AG and PAT
refer to agent and patient participants, respectively. These terms are used here in a
rather loosely defined sense and they cover all possible instigators (AG) and targets
(PAT) irrespective of whether these terms are semantically appropriate. The simple
arrow (÷) indicates the energy flow from agent to patient. Whenever necessary,
less-of-agents (like ‘forces’) and less-of-patients (e.g. stimuli) are distinguished
from typical instances by putting them in citation marks (i.e. “AG” and “PAT” or
‘AG’ and ‘PAT’, labels in single citation marks are more typical agents or patients
than those in double ones). Round brackets refer to semantic omission or
irrelevance/optionality of a participant. If the arrow appears in brackets, the energy
flow is viewed as less effective or concrete. If a participant appears inside round
brackets, it is regarded as being semantically irrelevant, peripheral or less integrated
into an event. Square brackets are used in two different ways. In the case of de-
transitive alternations, they refer to structural optionality (this use is restricted to de-
transitive alternations). Square brackets are elsewhere used to show that an event as
a whole may affect a participant. This is the case, for example, in 12 in which the
event as a whole affects the agent. The event is obviously not considered optional
in this case. We have opted for using square brackets in these kinds of case, since
round brackets refer to semantic optionality or irrelevance and is not intended in
466

cases like 12. In the case of both round and square brackets, single brackets refer to
a lower degree of optionality. Participants in double brackets are very irrelevant or
structurally completely omitted (in 17b and 18b, we have opted for using double
square brackets in order to distinguish agentless passives and patientless
antipassives from constructions in which participants are completely omitted or are
viewed as very peripheral). For convenience, we only refer to examples in the
section 5.3., even if the given alternations have been illustrated elsewhere as well.
The obvious ‘gaps’ in the numbering are due to the fact that some of the examples
illustrate other aspects and cannot consequently be classified as belonging to none
of the groups presented. The numbered cases illustrate the result of alternations. The
underlying structure is not numbered.

Minus-transitivizing alternations

Underlying structure AG ÷ PAT

1. ÷ PAT (Anticausative/intransitive inchoative)

Exx. 946-947, 954-957

(AG ÷ (Intr. agentive))

2. ((AG)) ÷ PAT (Resultative)

Exx. 948-957

3. AG ÷ PAT=AG (Reflexive)
Exx.958-971
3b. AG ÷ PAT ( ÷) AG (Indirect reflexive)
Exx.972-975

4. “AG” ( ÷) “PAT” (Dynamic)

Exx. 976-981

5. “AG” ( ÷) “PAT” (Stative)

Exx. 982-988

6. AG (÷ PAT) (Potential / desiderative constructions, habitual

events, future splits)
Exx. 989-1016

7. (AG ÷ PAT) (Possiblitive mood)

Exx. 1017-1022
 467

8. AG ÷ ‘PAT’ (Less successfully completed events, events

involving volitional patients)
Exx. 1023-1038, 1043-1048

9. AG ÷ ((PAT)) (Unsuccessful events, patient simply targeted

at, but not affected)
Exx. 1039-1042

10. “AG” ÷ PAT (Forces as agent)

Exx. 1052-1059
10b. AG ÷ PAT / “AG” ÷ PAT (Optional marking of the former)
Exx. 1060-1067

11. INSTR ÷ PAT (Instruments as agents)

Exx. 1068-1080

12. [AG ÷ PAT] ÷AG (Affected agent)

Exx. 1081-1086

13. ‘AG’ ÷ PAT (Forced/ impulsive/ ’unconscious’/ partial/

unintentional agents)
Exx. 1087-1106, 1122-1129

14. AG ( ÷) PAT (Agent not controlling the completion of events)

Exx. 1107-111

15. “AG” ÷ PAT (IAC’s)

Exx. 1112-1121

16. ‘AG’ ø ‘PAT’ (Reciprocal)

Exx. 1141-1150

De-transitive alternations

Underlying structure AG ÷ PAT

17. [AG] ÷ PAT (Passive)

Exx. 1151-1154, 1157-1159, 1168-1169, 1172
17b. [[AG]] ÷ PAT (Agentless passive)
Exx. 1155-1156

18. AG ÷ [PAT] (Antipassive)

Exx. 1160-1163, 1166-1167, 1171
18b. AG ÷ [[PAT]] (Patientless antipassive)
Exx. 1164-1165

19. AGPAT ÷ (Incorporation)

Exx. 1173-1178
468

20. ‘AG’ ÷ ‘PAT’ (De-transitives without argument omission)

Exx. 1179-1192

Transitivizing alternations

Underlying structure ÷ PAT

21. AG ÷ PAT (Typical causatives)

Exx. 1193-1198

22. AG ÷ ‘PAT’ (Causatives involving animate patients)

Exx. 1199-1206

23. AG ( ÷) ‘PAT’ (Directive causation)

23b. AG (( ÷)) ‘PAT’ (Less direct directive causation)
Exx. 1207-1218 (both illustrated here)

24. “AG” ( ÷) PAT (Indirect causation)

Exx. 1225-1236

25. “AG” ÷ PAT (Accidental causation)

Exx. 1237-1244

Underlying structure AG/PAT ÷

26. PAT ÷ PAT (Extended patientive intransitives)

Exx. 1245-1247

27. AG ÷ PAT (Extended agentive intransitives, type 1)

Exx. 1250-1255

27b. AG (÷ PAT) (Extended agentive intransitives, type 2)

Exx. 1248-1249, 1256-1261

28. [AG/PAT ÷] ( ÷) ‘PAT’ (Malefactive/benefactive alternations)

Exx. 1262-1275

29. AG (÷ OBJ) (Cognate object constructions)

Exx. 1276-1282
 469

30. AG (÷ X) (Applicatives)
Exx. 1283-1294

Underlying structure “AG” ÷ “PAT”

31. AG ÷ PAT (Transitivization of less transitive two-participant

events)
Exx. 1295-1302

Underlying structure AG (÷ ‘PAT’)

32. AG ÷ PAT (Transitivization of extended intransitives)

Exx. 1303-1306

Underlying structure AG ÷ PAT

33. AG ÷ PAT (Transitive events involving highly affected patients)

Exx. 1307-1316

Underlying structure AG ÷ PAT (÷ ‘PAT’)

34. [AG ÷ PAT] ÷ PAT (Transitive events involving both a directly and an
indirectly affected participant)
Exx. 1317-1321

Underlying structure “AG” ÷ PAT

35. AG ÷ PAT (Agentivization)

Exx. 1322-1326

Indirect alternations

Underlying structure AG ÷ PAT

36. AG ÷ PAT (Structure)

“AG” ( ÷) “PAT” (Semantics)
470

Exx. 1327-1388

Structurally motivated alternations

Underlying structure AG ÷ PAT

37. AG ÷ PAT (Semantics)

“AG” ÷ “PAT” (Structure)
Exx. 1389-1422, 352-356
6. SUMMARY

The present study has discussed the notion of linguistic transitivity from many
different perspectives. We hope to have shown that transitivity is cross-linguistically
a vast phenomenon, because of which transitivity definitions that focus on one
single aspect are always somehow insufficient. For example, definitions here
labelled as pragmatic are only concerned with aspects of definiteness, which does
make a very significant contribution to high structural transitivity, but which alone
explains only a small percentage of all possible transitivity alternations. Semantic
definitions proposed, for example, by Givón and Lazard (cf. ch. 2) are typologically
much more applicable in this respect, even if they also have some inadequacies.
Transitivity is clearly a combination of semantic, pragmatic and purely structural
features, all of which should be taken into account, if our goal is to study transitivity
exhaustively. In this chapter, we briefly summarize some of the most important
findings of our study and also discuss some topics for future research.
The primary goal of the presentation has been to explicitly show what kinds
of semantic (or ontological), pragmatic and also primarily structural features
contribute to transitivity. In the study we have proceeded from semantics to syntax.
We adopted this approach, since it is the best way of showing which aspects of
semantic (or ontological) transitivity are most relevant for the notion of transitivity
and thus makes it possible for us to achieve our goals. This has not only enabled us
to focus on the typologically relevant (semantic) features, but also made it possible
to take a look at how these are expressed in structurally diverse languages (i.e. to
study structural transitivity in detail). Another possibility would have been to study
the distribution and functions of different alternations. This means that we define
the notion of transitivity alternations structurally and study what kinds of
alternations are related to them. In this case, the starting point would be an
individual language. The results would have probably been in many ways similar to
those of the present study. However, the main problem in this of approach in studies
like ours is that we may have (unintentionally) ignored certain alternations. As
explicitly shown above, not only passives, antipassives and causatives contribute to
linguistic transitivity, but other kinds of structure are of significance as well. These
would have been ignored, if we had focussed on ‘traditional’ transitivity alternations
and their functions. Furthermore, we would not have been able to present a detailed
472

structural typology, since the structural features under study would have been
defined in advance. This is not to say that the latter kind of approach would be
completely useless in studies of transitivity. Quite the contrary. It can provide us
valuable information on the similarity relations between different alternations. We
already know, for example, that reflexives are used as functional passives in many
languages. By taking a look at what kinds of semantic alternations are usually
related to what structural alternations, we can go deeper into the realms of
transitivity. This clearly lies outside the scope of the present, rather bulky study, but
we look forward to future studies of this nature.
As noted above, transitivity is best defined as a combination of semantic,
pragmatic and primarily structural features all of which make a significant
contribution to the notion of transitivity. The primary starting point of our study was
semantic transitivity and its manifestation in languages. The most important single
feature of semantic transitivity is the involvement of distinct agent and patient
participants in events. We have been unable to reveal anything that would contradict
this and would necessitate modifications of this basic idea (and it is rather safe to
say that future studies will neither). Agent and patient are both integral parts of
typical transitive events and neither of them can be omitted without decreasing
transitivity. Typically, only events that involve both an agent and a patient to begin
with can receive a transitive morphosyntactic coding. Events involving participants
that cannot be labelled as typical agents or patients are less transitive and are
encoded less transitively. What makes the picture messier and thus more interesting
is that the expression or marking of arguments is conditioned by other factors as
well. A typical example of this is provided by the quite frequent omission of
arguments referring to indefinite participants that need not be stated explicitly.
Semantics and pragmatics of transitivity can be viewed as conditioning factors of
transitivity, whereas structure only reflect changes in these two features. However,
as has been shown, certain structural features of languages can alone be responsible
for transitivity alternations. Semantically and pragmatically motivated transitivity
alternations differ from each other in that the former have a clear non-linguistic
counterpart (e.g. events involving one participant and those involving two are
conceptually clearly different from each other), whereas the latter are typically
observed only via structural differences (non-linguistically a single entity can be a
 473

definite or an indefinite patient, for example).

Semantic transitivity is dividable into two completely independent
components labelled here as transitive valence and semantic transitivity. Semantic
transitivity can further be divided into ontological and conceptual facets. As has
been shown above, both make a very significant contribution to marking of
transitivity. Since we view the involvement of two independent participants as a
prerequisite of transitivity in our study, the contribution of transitive valence to
transitivity is viewed as more important here. Only events that involve two
participants can receive a transitive coding (obvious exceptions are provided by
reflexives and cognate object constructions). However, as has been shown naby
times throughout the present study, the mere existence of two participants is very
far from being a sufficient criterion of high linguistic transitivity in many languages,
but the profiled events have to correspond to the definition of a typical transitive
event in order to be encoded transitively. In this respect, both ontological and
conceptual transitivity features are more or less equally important (cf. below). This
means, for example, that an objectively salient change in affectedness parameter as
well as a subjective change in agency can both result in a transitivity alternation.
Languages differ crucially from each other in what kinds of event can be coded
transitively. In languages that are most sensitive to semantic transitivity, only highly
transitive events involving a typical agent and a typical patient are encoded by
transitive constructions. In others, the mere transitive valence suffices. Moreover,
there are also languages that seem to depend on the mere number of participants
without any reference to semantic roles. The latter is typical of analytic languages,
since they lack morphological means to distinguish between semantically different
arguments. In this case, however, the ‘freedom of choice’ is lacking. More genuine
examples of this are illustrated by languages like Creek that have ‘opted’ for this
kind of marking despite having sufficient morphological means of distinguishing
various (‘non-subject’) arguments explicitly. In the latter case, the notion of
semantic transitivity is irrelevant.
As regards the binary marking of basic transitivity or intransitivity (i.e.
transitivity that can be inferred from the semantics of verbs alone), high transitivity
is without exception more explicitly marked than intransitivity. We have not come
across languages in which intransitivity is explicitly marked, whereas transitivity is
474

unmarked. The most likely candidates are languages like Indonesian and Salish in
which both basic transitivity and basic intransitivity are signalled by distinct verbal
markers. The structural markedness of transitive clauses in comparison with
intransitive ones is very natural, since transitive clauses involve two arguments,
which naturally means that the number of relevant morphological means necessary
is higher than in the case of intransitive clauses. For example, instead of one, two
arguments may be cross-referenced on the verb, or the explicit reference to two
participants makes it necessary to disambiguate the semantic role assignment in
which case one of the arguments has to be explicitly marked in a non-core case. The
more explicit marking of intransitivity would clearly violate the economy principle
of language use and is consequently, as far as we know, not attested.
The principle presented above holds true only for marking of basic
(in)transitivity. As a whole, explicit expression is not restricted to high transitivity,
but there are, as generally known, also cases in which decreased transitivity
represents the marked case. Obvious examples of this are presented by
anticausatives and reflexives that are quite frequent cross-linguistically (as
morphologically distinct constructions). Cases in which high transitivity is
unmarked and decreased transitivity marked exemplify intransitivizing (or de-
transitivizing) alternations, while the opposite illustrates transitivizing alternations.
In distinction from the marking of basic (in)transitivity, these illustrate cases in
which some individual transitivity features are affected. Events like ‘he painted the
house’ and ‘he ran’ not only differ from each other in many relevant transitivity
features, but also the basic nature of the events is crucially different. On the other
hand, events like ‘he killed the man’ vs. ‘the man died’ can be regarded as transitive
and intransitive realizations of a single basic event that results in the death of a man.
The basic semantics of the verb employed in the description of the given event
allows two different interpretations that have to be explicitly distinguished from
each other in order to avoid misinterpretations. Hence, explicit marking of
(in)transitivity is necessary. In case the former (‘he killed the man’) is marked, we
are dealing with a transitivizing alternation. Typically (cf. e.g. Nichols 1982,
Haspelmath 1993), transitivization or intransitivization is restricted to cases in
which these operations are morphologically signalled on the verb. For example,
Nichols (1982: 457) has divided languages into ‘fundamentally transitive’ and
 475

‘fundamentally intransitive’ based on whether transitivizing or intransitivizing

verbal operations are more typical of a given language. If transitivizing alternations
are predominant, the language in question is ‘fundamentally intransitive’, whereas
the language is classified as ‘fundamentally transitive’ in the opposite case. In the
former case, intransitivity is more typical of verbs, which implies that verbs have to
be transitivized in case used transitively. The division proposed by Nichols is very
useful, since it underlines the differential expression of changes in basic transitivity
and in individual transitivity aspects (the latter covers all the cases in which events
obviously share common features). These two semantic alternations differ crucially
from each other in their morphosyntax. Alternations due to differences in inherent
transitivity are extremely rarely (if ever) expressed by manipulating the verb
morphology. Put concretely, this means that differences like ‘he ate the bread’, ‘he
saw me’ and ‘he ran’ are not expressed by modifying the verb morphology, but only
the number or marking of arguments is affected (cases in which transitivity of the
clause requires certain kind of morphological marking of the verb are excluded,
since the conditioning factor is structural transitivity instead of semantic one). On
the other hand, alternations like causatives, anticausatives, reflexives, passives and
antipassives are frequently marked on the verb as well. This kind of distribution is
predicted, if we take the economy principle seriously. Differences between events
like ‘he killed the man’ and ‘he saw the man’ can be readily inferred from the verb
semantics, which makes explicit marking on the verb redundant. In case a given
language restricts the use of basic transitive pattern to highly transitive events, the
case marking of arguments may be manipulated. On the other hand, obvious
changes in the valency usually have to be signalled on the verb, since it is not
possible to infer the intended meaning without explicit linguistic marking. As a very
broad generalization holds that the semantic markedness correlates with structural
markedness. This means that verbs (and also constructions in general) are unmarked
or less marked in case the semantics of the verb coincides with the inherent
semantics of the verb. In case the semantic transitivity of a given event differs
sufficiently from that inhering in the verb, an explicit marking is necessary. Because
of this, it is equally possible for both high (or increased) and low (or decreased)
transitivity to illustrate the marked case. There are no valid arguments that either of
these should be cross-linguistically preferred. Our findings coincide with this.
476

A further relevant distinction related to structural transitivity has to do with

directness of marking. Transitivity can be manifested either directly or indirectly.
These mechanisms are not mutually exclusive and in languages with indirect
marking (cf. (120)-(131)), direct marking is employed as well. Direct marking is
what most of us are familiar with and it can justly be labelled as the norm. Direct
marking refers to marking related to arguments and to verb morphology. Direct
marking is responsible for how clauses are interpreted. Indirect marking, on the
other hand, refers to structural aspects of clauses that are sensitive to transitivity, but
which do not qualify as instances of direct marking. Typical examples of this are
presented by transitivity concord of adverbials and other adjunct-like constituents.
In these cases, adverbials have to match the transitivity of clauses. Deviations are
deemed ungrammatical. Indirect marking is never the sole indicator of semantic
transitivity (at least no examples of this kind are known to us), even if there are
cases in which indirect marking is the only structural difference between two clauses
(cf. e.g. (124)-(127)), but also in this case the semantics of the verb is responsible
for our interpretation of the event. Differently from direct marking, indirect marking
is usually strictly binary (in light of examples in our data). For example, in the case
of adverbial suffix concord (Loos 1999: 239), the form of the adverbial suffixes is
manipulated according to whether the event in question is transitive or intransitive
(ditransitivity seems to be treated similarly to transitivity). Different degrees of
semantic transitivity or the precise motivation of transitive vs. less transitive
marking are irrelevant as regards the indirect marking. One of the reasons of the
secondary nature of indirect marking is that adverbials or adjuncts are always
optional constituents in clauses, because of which they could also in theory express
changes in transitivity only if adverbials are present. For the obviously secondary
nature of indirect marking, we have focussed on direct marking in our study. Despite
its obvious secondary nature, indirect marking is, however, a very interesting facet
of transitivity, since it provides us with evidence for the significance of purely
structural transitivity. Consequently, it would not make justice to the notion of
transitivity as a whole to ignore this aspect.
Morphosyntactic marking of transitivity can be conditioned by ontology of
events, semantics of verbs, semantics or pragmatics associated with situations, in
addition to which certain structural features make a significant contribution as well.
 477

In the first case, the relation of structural and semantic transitivity is rather direct.
If the denoted event is transitive, so is the structure employed in its description.
Ontology of events can be viewed as the basis for all marking of transitivity. This
is, among other things, reflected in the highly transitive encoding of highly transitive
events. These kinds of event always have the possibility of appearing in transitive
frames, whereas obviously less transitive ones have not. Typically, ontological
transitivity coincides with the semantics of verbs. This means that the semantics of
a verb coincides directly with the ontological transitivity of an event. For example,
the English verb paint requires that there be an agent acting volitionally and a
patient affected by the event. At the level of morphosyntax, this is usually reflected
in that the clause employed in the description of this kind of event has an Agent and
a Patient argument. In case the marking is conditioned by the ontology of events, it
is irrelevant how the events are referred to. Typically, all possible verbs employed
to denote a certain event behave identically as regards the marking of transitivity.
Moreover, there are verbs the semantics of which matches less than ideally with the
ontology of events. In this case, two or more semantically very similar, yet
somewhat different verbs are used to profile one basic event. There might (but does
not have to) be certain subtle differences that result in differences in the expression
of basic transitivity or in the application of transitivity alternations (examples were
presented in (394)-(401)). Since the basic event is the same, we may conclude that
the semantics of verbs (or verb classes) is responsible for the differences attested.
Furthermore, not only the basic ontological transitivity should be taken account of,
but also situational transitivity of events is important. Transitivity of events is not
a constant notion, but it varies. This refers here to cases in which the transitive
valence is not affected, but the changes are related to various individual aspects of
agency or affectedness parameters. For example, an agent may be forced to carry out
a transitive action, which naturally affects the agency adversely. Expectedness is
very important in this respect. This means that the actual degree of transitivity does
not coincide with our expectations, which makes linguistic marking unnecessary.
Pragmatics also contributes in a significant way to structural transitivity.
Definiteness of arguments is the most relevant facet of pragmatic transitivity. In
numerous languages, indefinite arguments are simply omitted, which naturally
reduces transitivity. Another aspect related to definiteness is the differential object
478

marking. In languages in which this relevant, the Patient of basic transitive clauses
has to be definite. In addition to semantics and pragmatics, certain structural
properties of clauses also affect transitivity. The verb morphology is especially
important in this respect. Typical examples are provided by the expression and
marking of arguments in passives and antipassives. In some languages, passives
(e.g. Finnish, Ute and Udi) or antipassives (e.g. Hunzib and Toqabaqita) derived
from transitive clauses are obligatorily completely de-transitive constructions. As
opposed to anticausatives and agentive intransitives, the structural intransitivity (i.e.
de-transitivity) is not conditioned by semantics, but by verb morphology, which
makes it impossible to express the Agent or the Patient explicitly. The opposite is
illustrated by some applicatives in which two arguments have to be explicitly
expressed, even if the event denoted is intransitive (cf. (1285)-(1290)). In the latter
two cases, semantic and structural transitivity often disagree, since arguments are
added or omitted without any obvious ontological motivation. In causatives and
anticausatives, the expression of arguments is required by both semantics and
structure, since the verb morphology makes it impossible or obligatory to refer to
the agent, while, on the other hand, the marking of the verb is conditioned by
semantics. It would be very interesting to study the differences between verbs whose
transitivity is only semantic, and cases in which the marking of arguments is
conditioned by structure as well. Intuitively, it would seem that in the latter kinds
of case, the expression of arguments is more obligatory, but we do not wish to
propose this analysis before taking a closer look at this.
Expression of transitivity is highly language-specific. First of all, languages
differ vastly from each other in what kinds of semantic change result in a transitivity
alternation. Some languages ‘tolerate’ far greater changes than others without
manipulating the morphosyntax of clauses. It is typical of these languages that the
basic transitive pattern can be employed also in the description of events the
transitivity of which is considerably lower than that of typical transitive events
(according to the view adopted by Drossard (1991:413) these languages are ‘more
transitive’ than others). Furthermore, some languages simply have a larger number
of different kinds of transitivity alternation than others. A feature of semantic
transitivity that is relevant in language A can be totally irrelevant in language B.
This was illustrated in light of Finnish and German in (370)-(374). In Finnish, the
 479

negation can be regarded as a genuine transitivity alternation, whereas in German

it is not sensitive to transitivity in any way. Languages also diverge in what
morphosyntactic mechanisms are employed in the expression of alternations. As
noted above (see 5.1.), alternations expressed by manipulating the number or case
marking of arguments are the most obvious examples of transitivity alternations.
This is due to the fact that these changes are in many languages related to the
expression of lower degrees of transitivity in general. Alternations expressed by
manipulating the verb morphology are much less obvious in this respect. They
cannot be automatically labelled as transitivity alternations. We cannot usually
justify our claims by referring to other constructions the transitivity of which is
clearly lower. Typically, only semantics can aid us in distinguishing genuine
transitivity alternations from false ones. This results in a further problem. In order
to be able to take account of genuine transitivity alternations only, we should be able
to define the notion of transitivity accurately on the basis of semantics. In this case,
every morphosyntactic change caused by a change in some facet of transitivity could
be labelled as a genuine transitivity alternation. This is, however, not possible, since
other aspects affect the marking of transitivity as well. Since the starting point in our
typology is the semantics, it seems unjust to disregard transitivity alternations that
are marked solely by manipulating verb morphology. These also reflect changes in
semantic transitivity. Ignoring these kinds of alternation seems particularly
unmotivated, if the differences in transitivity between two clauses are very obvious.
One option is to use cross-linguistic evidence here. This means that, if a certain
change in semantic transitivity is typologically very relevant and produces a genuine
transitivity alternation in a great number of languages, we are justified to label less
obvious cases as transitivity alternations as well. The structural manifestation is
simply less typical. This, perhaps somewhat unorthodox approach, is applicable only
if the semantic differences between two clauses are very obvious. If an alternation
is attested in two languages out of 600, it seems less justified to label semantically
similar, yet structurally less obvious changes in the structure of clauses as
transitivity alternations of any kind.
Above, we briefly summarized the discussion of alternations caused by the
fact languages express different alternations using different morphosyntactic
mechanisms. The effects of language structure manifest themselves also in other
480

ways. One obvious aspect of this is the morphological nature of a given language.
This refers to the simple established fact that analytic languages do not mark
arguments for case nor do they manipulate the verb morphologically, which
naturally reduces the number of mechanisms available for transitivity expression.
Consequently, it does not surprise us that analytic languages are in general less
susceptible to the expression of transitivity than languages with a rich morphology.
For example, in analytic languages with no morphological means of expressing
location, clauses like ‘he sat in a chair’ may also appear in the basic transitive frame
(cf. e.g. (11) from Kammu). The ‘freedom of choice’ is lacking, since it is simply
impossible for a language to express a variety of transitivity alternations explicitly.
Cases in which a language has the morphological means to express certain
alternations explicitly without doing this are of greater interest. Another relevant
aspect of language structure that clearly affects the nature of transitivity alternations
expressed in a given language is presented by differences in basic argument pattern.
As briefly discussed in 5.2.1. (see exx. (431)-(438)), morphologically distinct
passives are more frequent in predominantly accusative languages, whereas
antipassives dominate in predominantly ergative languages. Also in other respects,
accusative languages express changes in affectedness more easily than ergative
languages that are more sensitive to agency. The rationale behind the alternations
is the same, but the obviousness depends on the argument marking pattern of a
given language.
The notions of basic transitive (and intransitive) clauses and transitivity
alternations are intimately related to each other. We can learn a great deal about the
basic transitive clause by studying transitivity alternations in detail. Any change in
transitivity expressed explicitly by manipulating the structure of the basic transitive
clause can be considered relevant for the notion of high transitivity. The lack of the
feature in question reduces the transitivity of events and clauses, which means that
feature in question is relevant for high transitivity and should therefore be taken
account of when defining the notion ‘basic transitive clause’. Every feature of this
kind can be considered significant in typological definitions. The most typical
semantic features associated with the given concept are related to agency and
affectedness parameters. A basic transitive event is a single, independent event with
two distinct participants one of which can be regarded as a typical agent, whereas
 481

the other is directly and in a salient way affected by it. The energy flow is clearly
unilateral and runs from agent to patient. The agent is acting volitionally,
intentionally (including consciousness) and is controlling the given event.
Furthermore, the agent must not be an affected entity, but the effect of the event as
a whole manifests itself on the patient. The patient of typical transitive events is a
passive entity that simply registers the effect of the event. It is incapable of
controlling the event and its participation cannot be regarded as volitional or
intentional. The inventory of roles in basic transitive events has to be unequivocal
and both participants bear one single role only with no overlaps. Moreover, basic
transitive events have to be successfully completed, concrete events for the
occurrence of which we need to have some evidence. As regards the identity of
participants in transitive events, both participants have to be definite in nature and
need to have an identifiable non-linguistic referent. Grammatically, basic transitive
events are independent clauses with a non-derived verb morphology. Any deviation
from this can result in a transitivity alternation, which was illustrated in section 5.3.
Structurally, the distinction between basic transitive and intransitive clauses
is based on the number of arguments, their case marking, different aspects of verbal
cross-reference and the verb morphology. In our typology, we distinguished two
main types of basic transitive clause that were labelled as quantitative and
qualitative types. In the former case, basic transitive and intransitive clauses are
distinguished from each other only on the basis of the number of mechanisms
employed. In quantitative cases, there are no morphosyntactic mechanisms in
transitive clauses that could not be parts of intransitive clauses as well. A typical
example of this is provided by (analytic) languages in which the only difference
between basic intransitive and transitive clauses is the number of arguments. In
qualitative types, certain structural element(s) is (are) attested only in transitive
clauses, which per se is a sufficient indicator of transitivity. The most salient feature
of this kind is the case marking of arguments. In the majority languages, the sole
argument of intransitive clauses is zero marked, whereas at least one of the
arguments in transitive clauses is marked distinctively. The explicitly marked
arguments can appear only in transitive clauses, which means that, for example, the
accusative marking of Patient suffices to distinguish between intransitive and
transitive clauses. Depending on the language, one or both arguments of transitive
482

clauses are marked distinctively from the S argument. Additionally, verb agreement
is in many languages qualitative and enables an unambiguous distinction between
transitive and intransitive clauses. In some languages, the agreement as a whole is
restricted to transitive clauses. A typical example is provided by Barai. In other
languages, on the other hand, the nature of agreement is sensitive to transitivity of
clauses. The agreement may be retained (as, e.g. in Chamorro and Dolakh~ New~r),
but it may have shifted. For example, in Dolakh~ New~r the verb agrees (by default)
with the zero marked argument in intransitive clauses, whereas the agreement is
with the ergatively marked Agent in transitive clauses. In the third kind of case, the
agreement is constant, but the form of the agreement affix varies. An example is
presented by Kolyma Yukaghir. One final possibility is that the number of
arguments agreed with changes, but differently from quantitative types, the nature
of agreement is different from that in intransitive clauses.
Our typology of basic transitive transitive clauses was based on the structural
differences between basic intransitive and transitive clauses. Semantically, the
typology emphasizes differences between distinct events. The typology of
transitivity alternations, on the other hand, focusses on minor differences between
different instances of the same basic events. The notion of transitivity alternations
comprises here all the cases in which an obvious change in some aspect of
transitivity is somehow expressed morphosyntactically. Depending on the effect the
alternations have on the structure of clauses, transitivity alternations can be divided
into transitivizing, intransitivizing and transitivity rearranging alternations.
Furthermore, the alternations differ from each other in whether they affect the
number of participants, number of arguments or only some minor semantic and
structural features of events and clauses. Both of these threefold divisions are very
relevant for the typology of transitivity alternations and they cover every possible
alternations there is. The three subclasses of the latter are completely independent
of each other, which means that is possible for an alternation to affect the number
of arguments without having any consequences for the number of participants and
vice versa (e.g. passives exemplify this). The former were here labelled as de-
transitivizing alternations for the purpose of explicitly distinguishing them from
semantically motivated intransitivizing alternations. Structurally, the most relevant
features related to the expression of transitivity alternations are also the number and
 483

case marking of arguments along with different aspects of verb morphology.

Differently from basic transitivity, verb morphology makes a far more significant
contribution to the expression of transitivity alternations. Usually, only alternations
with a manipulated verb morphology are included in studies of transitivity
alternations. An illustrative example of this is presented by definitions of typical
passives (cf. e.g. Siewierska 1984: 2, Shibatani 1985: 837) in which the verb
morphology is an integral feature of passives. As has been noted, this kind of
distribution of marking is very natural. Transitive vs. intransitive marking of verbs
in basic cases is very redundant, since it does not reveal anything about the nature
of events that is not inferable from the semantics of the verb alone. On the other
hand, in the case of transitivity alternations, the semantics of events changes in ways
that have to be explicitly expressed. One of the options is to manipulate verb
morphology, which is quite frequently attested.
Transitivity alternations can be motivated semantically or structurally and they
can be expressed either directly or indirectly. The former is more typical in both
cases. On the basis of these two properties, transitivity alternations were divided into
three main types (the combination of structural motivation/indirect expression is not
possible). Typical transitivity alternations are conditioned by different aspects of
agency and affectedness parameters, in addition to which features of dynamicity and
concreteness of events also make a very significant contribution. In our typology,
different alternations due to changes in agency seem to outnumber those due to other
aspects, which likely follows from the multilayered notion of agency. Agency
consists of many different aspects and a change in each of these can result in a
transitivity alternation. These may all be expressed by a single alternation, but
typologically the number of different agency-conditioned alternations is quite high.
As regards the structural realization of transitivity alternations, they are usually
expressed directly employing one of the mechanisms noted above. This means, for
example, that a change in the agency is expressed linguistically by manipulating the
marking of Agent. The other possibility is that the semantic changes are reflected
indirectly. In this case, the structure of basic clauses is not necessarily affected in
any way, but the semantic changes are reflected in other ways. There are two main
types of this. First, differences in semantic transitivity can result in optional changes
in the basic structure. Typical examples were presented in (1060)-(1067) in which
484

a lower degree of agency makes it possible for a clause to appear in two case frames
instead of one. Second, the semantic changes can affect the application or structure
of transitivity alternations. For example, in some languages, passivization is possible
only if the non-derived clause is sufficiently transitive. In other languages, the
derivation of passives or antipassives is possible, but the structure of the derived
constructions varies. Examples include antipassives of some Australian languages
in which the form of the Patient of antipassives is sensitive to definiteness (or
humanness). In both types briefly discussed above, we can show that the underlying
motivation is some semantic (or pragmatic) aspect. Moreover, there are alternations
that are not motivated by semantics or pragmatics . These were here labelled as
structurally conditioned alternations. Any change that resembles a semantically
induced alternation (even without any semantic motivation) is included in the latter
type.
Even if the present study illustrates our final word on the typology of
transitivity, there are many aspects worth studying further. Due to the vastness of the
study, some individual aspects have only been touched upon and further studies are
needed to shed more light on the topic. Hence, we end by discussing some of these
aspects and by hoping that others will find some of these aspects worth an
independent study. We believe to have discussed the semantic rationale behind
alternations sufficiently. This means that future studies should not reveal anything
radically new in this respect. Also the structural typology both of basic transitive
clauses and transitivity alternations should cover by far most of the cases attested
in the languages of the world. Only some minor modifications might be needed. The
two features, a closer study on which is most sorely needed, are indirect alternations
and structurally motivated alternations. Due to insufficient discussion of these two
alternations in the data, we have been able to discuss these two extremely interesting
alternations types only in a very preliminary way. A more detailed study may reveal
much more about how syntax and semantics cooperate to express transitivity. The
ways in which indirect alternations are reflected (e.g. which alternations or
morphosyntactic operations are affected) could tell us very much about what is
important for structural transitivity cross-linguistically. An interesting individual
aspect related to indirect alternations is presented by restrictions on antipassivization
that are extremely rarely discussed. The effects of differences in motivation of
 485

transitivity on the application of transitivity alternations are rather closely related to

indirect alternations. We were able to discuss this only briefly and presented only
some very general guidelines that should be made more accurate on the basis of
more detailed studies. The three differently motivated types of transitive events
diverge crucially as regards how intimately participants are integrated in events
which naturally has consequences for which participants can be omitted. Closer
studies on indirect marking could also shed more light on the structural typology of
transitivity.134 Also the effects of the structure of languages (including basic
argument marking pattern and morphological type) on the expression of transitivity
are worth examining further. As briefly noted, accusative languages express changes
in the affectedness parameter more readily than ergative languages, which, on the
other hand, are more sensitive to changes in agency. One reflection of this is the
predominance of morphological passives in accusative languages, whereas
antipassive is more typical of ergative languages. This claim should be verified by
a detailed typological study. A detailed study of effects of the morphological type
of a language on the expression of transitivity could aid us in understanding what
kinds of transitivity alternation can be viewed as the most basic. Analytic languages
are reluctant to express a variety of alternations explicitly due to structural
restrictions. Alternations that are expressed also in analytic languages may be the
most elementary ones typologically (for example, some kind of reflexive is attested
in all the languages, whereas alternations due to some minor aspects of agency are
clearly more language-specific). One possible topic for future research, which has
been (purposefully) fully ignored by us, is to study whether there are any
correlations between transitivity alternations. First of all, it would be of interest to
know, whether the existence of certain semantic alternation implies others. Second,
the structural implications associated with different alternations are also interesting.
This refers to the possible correlations of, for example, transitive reflexives with
explicitly marked anticausatives (this is a mere example and is not intended as a
possible correlation found).

134
Pilar Valenzuela has studied this aspect in more detail.
REFERENCES

Aikhenvald, Alexandra Y. 2000. Transitivity in Tariana. In: Dixon & Aikhenvald (eds.) 2000a, pp.
145-172.
Aikhenvald, Alexandra Y. & Dixon, Robert Malcolm W ard 1999. Other small families and isolates.
In Dixon & Aikhenvald (eds.) 1999, pp. 341-383.
Amberber, Mengistu. 2000. Valency-changing and valency-encoding devices in Amharic. In: Dixon
& Aikhenvald (eds.) 2000a, pp. 312-332.
Amritavalli, R. 1990. Case-marking choices for experiencer subjects. In Verma & Mohanan (eds.)
1990, pp. 285-295.
Andersen, Torben. 1988. Ergativity in Päri, a Nilotic OVS language. Lingua 75, pp. 289-324.
Anderson, Stephen R. 1976. On the notion of subject in ergative languages. In: Li (ed.) 1976, pp. 1-
23.
Austin, Peter K. 1981. A grammar of Diyari. Cambridge University Press.
Austin, Peter K.1982. Transitivity and cognate objects in Australian languages. In: Hopper and
Thompson (eds.) 1982, pp. 37-47.
Austin, Peter K.. 1995. Double case marking in Kanyara and Mantharta languages, W estern
Australia. In: Frans Plank (ed.), Agreement by Suffixaufnahme. Oxford University Press, pp.
363-379.
Austin, Peter K. 1997. Causatives and applicatives in Australian Aboriginal languages. In:
Matsumura, Kazuto & Hayasi, Tooru (eds.), The dative and related phenomena. Tokyo: Hituzi
Syobo, pp. 165-225.
Austin, Peter K. 1997b. Texts in the Martharta languages, W estern Australia. Tokyo: ILCAA.
Austin, Peter K.. 2000. Verbs, voice and valence in Sasak. In Peter Austin (ed.), W orking papers in
Sasak, Vol. 2. The University of Melbourne, pp. 5-24.
Baker, Mark C. 1988. Incorporation: A theory of grammatical function changing. Chicago:
University of Chicago Press.
Bani, E. & Klokeid T.J. 1976. Ergative switching in Kala Lagau Langgus. In: Sutton P. (ed.),
Languages of Cape York. Canberra: Australian Institute of Aboriginal Studies, pp. 269-
283.
Baratin, Marc. 1998. Aperçu sur la transitivité chez grammairiens anciens. In: Rousseau (ed.) 1998,
pp. 15-18.
Bausewein, Karin. 1990. Akkusativobjekt, Akkusativobjektsätze und Objektivsprädikate im
Deutschen. Untersuchungen zu ihrer Syntax und Semantik. Tübingen: Max Niemeyer Verlag.
Beale, T. 1974. A grammar of the Biri language of North Queensland. Typescript, Australian
National University, Canberra.
Bennett, David C. Bynon Theodora & Hewitt George B. (eds.) 1995. Subject, voice and ergativity.
Selected essays. University of London: School of Oriental and African Studies.
Bhat, D.N.S. & Ningomba, M.S. 1997. Manipuri grammar. München/Newcastle: Lincom Europa.
Bhatia, Tej K. 1993. Punjabi: A cognitive-descriptive grammar. London/New York: Routledge.
 487

Blake, Barry J. 1976. On ergativity and the notion of subject: Some Australian cases. In: Lingua 39,
pp. 281-300.
Blake, Barry J. 1977. Case marking is Australian languages. Canberra: Australian Institute of
Aboriginal Studies.
Blake, Barry J. 1979. A Kalkatungu grammar. Canberra: Canberra: Pacific Linguistics.
Blake, Barry J. 1979b. Pitta-Pitta. In Dixon & Blake (eds.) 1979, pp. 183-244. London: Crrom Helm.
Blake, Barry J. 1982. The absolutive: Its scope in English and Kalkatungu. In: Hopper and
Thompson (eds.) 1982, pp. 71-94.
Blake, Barry J. 1987. Australian Aboriginal grammar.
Blake, Barry J. 1994. Case. Cambridge University Press.
Blake, Barry J. & Breen, J.G. 1971. The Pitta-pitta dialects. Melbourne: Monash University.
Linguistic communications 4.
Blake, Barry J. & Mallinson, Graham. 1981. Language typology. Cross-linguistics studies in syntax.
Amsterdam/New York/Oxford: North-Holland Publishing Company.
Bowe, Heather J. 1990. Categories, constituents and constituent order in Pitjantjatjara. An
Aboriginal language of Australia. London/New York: Routledge.
Brainard, Sherri. 1997. Ergativity and grammatical relations in Karao. In: Givón (ed.) 1997, pp. 85-
154.
Breen, J.G. 1973. Bidyara and Gungabula grammar and vocabulary. Monash Linguistic
Communications.
Breen, J.G. 1976. Yandruwandha. In: Dixon (ed.) 1976, pp. 594-597.
Bril, Isebelle. 1994. Split ergative in the Nêlêmwâ language. Paper presented as 7 th International
Conference on Austronesian Linguistics, Leiden.
Brinker, K. 1971. Das Passiv im heutigen Deutsch I: Form und Funktion. Munich.
Bruce, Les. 1984. The Alambak language of Papua New Guinea (East Sepik). Canberra: Pacific
Linguistics.
Cain, Bruce D. & Gair James W . 2000. Dhivehi (Maldivian). München/Newcastle: Lincom Europa.
Camp, Elizabeth L. 1985. Split ergativity in Cavineña. In: International Journal of American
Linguistics 51, pp. 38-58.
Campbell, Lyle. 2000. Valency-changing derivations in K’iche’. In: Dixon & Aikhenvald (eds.)
2000, pp. 236-281.
Chapman, Shirley & Derbyshire, Desmond C. 1991. Paumarí. In: Derbyshire, Desmond C. & Pullum,
Geoffrey K (eds.), Handbook of Amazonian languages. Volume 3. Berlin/New York: Mouton
de Gruyter, pp. 161-352.
Charachidzé, G. 1981. Grammaire de la langue avar. Saint-Sulpize de Favières: Editions Jean-
Favart.
Chung, Sandra. 1976. An object-creating rule in Bahasa Indonesia. In: Linguistic Inquiry 7, pp. 41-
87.
Churchward, C.M. 1953. Tongan grammar. New York: Academic Press.
488

Colarusso, John. 1977. The languages of the North W est Caucasus. In: The languages of the Non-
Russian peoples of the Soviet Union. McMaster University and Canada Council, pp. 62-154.
Colarusso, John. 1989. East Circassian (Kabardian dialect). In: Hewitt, George (ed.), The Indigenous
languages of the Caucasus. Volume 2: The North West Caucasian languages. Delmar:
Caravan Books, pp. 263-355.
Colarusso, John. 1992. A grammar of the Kabardian language. University of Calgary Press.
Cole, Peter. 1983. The grammatical role of the causee in universal grammar. In: International
Journal of American Linguistics 49, 115-133.
Comrie, Bernard. 1973. The ergative: Variations on a theme. In: Lingua 32, pp. 239-253.
Comrie, Bernard. 1975. The antiergative: Finland’s answer to Basque. In: CLS 11, pp. 112-121.
Comrie, Bernard. 1976. The syntax of causative constructions. In Shibatani 1976a (ed.), pp. 261-312.
Comrie, Bernard. 1977. In defense of spontaneous demotion: the impersonal passive. In: Cole, Peter
& Saddock, Jerrod M. (eds.), Grammatical relations (Syntax and Semantics, Volume 8). New
York: Academic Press, pp. 47-58.
Comrie, Bernard. 1978. Ergativity. In: Lehmann, W infred P. (ed), Syntactic typology Sussex: The
Harvester Press, pp. 329-394.
Comrie, Bernard. 1982. Grammatical relations in Huichol. In: Hopper and Thompson (eds). 1982,
pp. 95-115.
Comrie, Bernard. 1985. Causative verb formation and other verb-deriving morphology. In: Timothy
Shopen (ed.) 1985, Language typology and syntactic description. Volume III: Grammatical
categories and the lexicon. Cambridge University Press, pp. 309-348.
Comrie, Bernard. 1988. Passive and voice. In: Shibatani (ed.) 1988, pp. 9-23.
Comrie, Bernard. 1989 [1981]. Language universals and linguistic typology. Chicago: University
of Chicago Press.
Comrie, Bernard. 1993. Some remarks on causatives and transitivity in Haruai. In: Comrie &
Polinsky (eds.) 1993, pp. 315-325.
Comrie, Bernard. 2000. Valency-changing derivations in Tsez. In: Dixon & Aikhenvald (eds.) 2000,
pp. 360-374.
Comrie, Bernard & Polinsky, M aria (eds.). 1993. Causatives and transitivity.
Amsterdam/Philadelphia: John Benjamins.
Cooreman, Ann M. 1987. Transitivity and discourse continuity in Chamorro narratives. Berlin/New
York/Amsterdam: Mouton de Gruyter.
Cooreman, Ann M. 1988. The antipassive in Chamorro: variations on the theme of transitivity. In:
Shibatani (ed) 1988, pp. 561-593.
Cooreman, Ann M . 1994. A functional typology of antipassives. In: Fox & Hopper (eds.) 1994, pp.
49-88.
Corston, S.H. 1996. Ergativity in Roviana, Solomon islands. Canberra: Pacific Linguistics.
Creissels, Denis. 2000. Typology. In: Heine & Nurse (eds.) 2000, pp. 231-258.
Croft, W illiam. 1990. Typology and universals. Cambridge University Press.
 489

Croft, W illiam. 1991. Syntactic categories and grammatical relations. Chicago and London:
University of Chicago Press.
Croft, W illiam. 1994. Voice: Beyond control and affectedness. In Fox & Hopper (eds.) 1994, pp. 89-
117.
Crowley, Terry. 1979. Yaygir. In Dixon & Blake (eds.) 1979, pp. 363-390.
Crowley, Terry. 1982. The Paamese language of Vanuatu. Canberra: Pacific Linguistics.
Crowley, Terry. 1983. Uradhi. In Dixon & Blake (eds.) 1983, pp. 307-430.
Crowley, Terry. 1987. Serial verbs in Paamese. In: Studies in Language 11, pp. 35-84.
Crowley, Terry. 1998. Ura. München/Newcastle: Lincom Europa.
Curnow, Timothy. 2001. Evidentiality and first person: an initial examination. Paper presented at
ALT IV, Santa Barbara, July 20, 2001 (handout).
Davidse, Kristin. 1992. Transitivity/ergativity: the Janus-headed grammar of actions and events. In:
Davies, Martin & Ravelli, Louise (eds.), Advances in systemic linguistics: recent theory and
practice. London: Pinter, pp. 105-135.
Davidse, Kristin. 1996. Ditransitivity and possession. In Hasan et al (eds.) 1996, pp. 85-144.
Davies, W illiam D. 1986. Choctaw verb agreement and universal grammar.
Dordrecht/Boston/Lancaster/Tokyo: D. Reidel.
Davies, W illiam. 1999. Madurese. München/Newcastle: Lincom Europa.
Davis, Karen. 1997. A grammar of the Hoava language, Western Solomons. PhD thesis, University
of Auckland.
Davis, Philip W . & Saunders, Ross. 1997. A grammar of Bella Coola. University of Montana
Occasional papers in Linguistics.
DeLancey, Scott. 1981. An interpretation of split ergativity and related patterns. In: Language 57,
pp. 626-657.
DeLancey, Scott. 1982. Aspect, transitivity and viewpoint. In: Hopper, Paul J. (ed.), Tense-aspect:
between semantics and pragmatics. Amsterdam/Philadelphia: John Benjamins, pp. 167-184.
DeLancey, Scott. 1984. Transitivity and case in Lhasa Tibetan. In BLS 1984, pp. 131-140.
DeLancey, Scott. 1984b. Notes on agentivity and causation. In: Studies in Language 8, pp. 181-213.
Dench, Alan Charles. 1991. Panyjima. In Dixon & Blake (eds.) 1991, pp. 124-243.
Dench, Alan Charles. 1995. Martuthunira. A language of the Pilbara region of Western Australia.
Canberra: Pacific Linguistics.
Dench, Alan Charles. 1998. Yingkarta. München/Newcastle: Lincom Europa.
Denwood, Philip 1999. Tibetan. Amsterdam/Philadelphia: John Benjamins.
Derbyshire, Desmond C. & Pullum, Geoffrey K. (eds.). 1986. Handbook of Amazonian languages.
Volume 1. Berlin/New York/Amsterdam: Mouton de Gruyter.
Desclés, Jean-Pierre. 1998. Transitivité sémantique, transitivité syntaxique. In: Rousseau (ed.) 1998,
pp. 161-180.
Desclés, Jean-Pierre & Guenthchéva, Zlatka. 1998. Causalité, causativité, transitivité. In Kulikov &
Vater (eds.) 1998, pp. 7-27.
490

De Vries, L. 1983. The passive-like constructions in Indonesian. In: Simon C. Dik (ed.), Advances
in functional grammar. Dordrecht: Foris Publications, pp. 155-173.
Dixon, Robert Malcolm W ard. 1972. The Dyirbal language of North Queensland. Cambridge
University Press.
Dixon, Robert Malcolm W ard (ed.). 1976. Grammatical categories in Australian languages. New
Jersey: Humanities Press.
Dixon, Robert Malcolm W ard. 1977. A grammar of Yidi . Cambridge University Press.
Dixon, Robert Malcolm W ard. 1979. Ergativity. In: Language 55, pp. 59-138.
Dixon, Robert Malcolm W ard. 1981. W argamay. In: Dixon & Blake (eds.) 1981, pp. 1-144.
Dixon, Robert Malcolm W ard (ed.). 1987. Studies in ergativity. Lingua 71.
Dixon, Robert Malcolm W ard. 1987b. Introduction. In: Dixon (ed.) 1987, pp. 1-16.
Dixon, Robert Malcolm W ard. 1988. A grammar of Boumaa Fijian. Chicago and London: University
of Chicago Press.
Dixon, Robert Malcolm W ard. 1991. Mbabaram. In: Dixon & Blake (eds.) 1991, pp. 348-402.
Dixon, Robert Malcolm W ard. 1994. Ergativity. Cambridge University Press.
Dixon, Robert Malcolm W ard. 2000. A typology of causatives: form, syntax and meaning. In: Dixon
& Aikhenvald (eds.) 2000a, pp. 30-83.
Dixon, Robert Malcolm W ard & Aikhenvald, Alexandra Y. 1997. A typology of argument-
determined constructions. In: Joan Bybee, John Haiman and Sandra A. Thompson (eds.),
Essays on language function and language type. Amsterdam/Philadelphia: John Benjamins,
pp. 71-113.
Dixon, Robert Malcolm W ard & Aikhenvald Alexandra Y. (eds.). 1999. The Amazonian languages.
Cambridge University Press.
Dixon, Robert Malcolm W ard & Aikhenvald Alexandra Y. (eds.). 2000a. Changing Valency. Case
Studies in transitivity. Cambridge University Press.
Dixon, Robert Malcolm W ard & Aikhenvald Alexandra Y. 2000b. Introduction. In: Dixon &
Aikhenvald (eds.) 2000a, pp. 1-29.
Dixon, Robert Malcolm W ard & Blake Barry J. (eds.). 1979. The handbook of Australian languages.
Volume I. Amsterdam/Philadelphia: John Benjamins.
Dixon, Robert Malcolm W ard & Blake, Barry J. (eds.). 1981. The handbook of Australian
languages. Volume II. Amsterdam/Philadelphia: John Benjamins.
Dixon, Robert Malcolm W ard & Blake, Barry J. (eds.). 1983. The handbook of Australian
languages. Volume III. Amsterdam/Philadelphia: John Benjamins.
Dixon, Robert Malcolm W ard & Blake Barry J. (eds.). 1991. The handbook of Australian languages.
Volume 4. The Aboriginal language of Melbourne and other grammatical sketches. Oxford
University Press.
Dixon, Robert Malcolm W ard & Blake, Barry J. (eds.). 2000. The handbook of Australian
languages. Volume 5. Oxford University Press.
Donaldson, Tamsin. 1980. Ngiyambaa. The language of the Wangaaybuwan. Cambridge University
 491

Press.
Donohue, Mark. 1997. Cognate objects, nonconfigurationality, (and other issues in ‘transitivity’ in
Kanum). Talk given at McGill Project Meeting, November 27, 1997.
Donohue, Mark. 1998. Transitivity in Tukang Besi. In: Studies in Language 22, pp. 83-111.
Donohue, Mark. 1998b. Intransitivity in Tukang Besi. Talk given at Stanford, April 7, 1998.
Donohue, Mark. 1999. A grammar of Tukang Besi. Berlin/New York: Mouton de Gruyter.
Donohue, Mark. 1999b. Warembori. München/Newcastle: Lincom Europa.
Donohue, Mark. 1999c. Syntactic roles vs. semantic roles: external possession in Tukang Besi. In
Payne & Barshi (eds.) 1999, pp. 373-401.
Dowty, David. 1991. Thematic protoroles and argument selection. In: Language 67, pp. 547-619.
Drossard, W erner. 1991a. Verbklassen. In: Seiler & Premper (eds.) 1991, pp. 150-182.
Drossard, W erner. 1991b. Transitivität (vs. Intransitivität) und Transitivierung (vs. Intransitivierung)
unter typologischem Aspekt. In: Seiler & Premper (eds.) 1991, pp 408-445.
Drossard, W erner. 1991c. Kasusmarkierung: Zur Zentralität und Peripherizität von Partizipanten. In
Seiler & Premper 1991 (eds.), pp. 446-481.
DuBois, Jack W . 1987. The discourse basis of ergativity. In: Language 63, pp.805-855.
Duchateau, Jean-Paul. 1998. Critères de transitivité. Essai de systématisation des critères, indices,
tests de transitivité. In Rousseau (ed.) 1998, pp.114-129.
Dunn, Michael John. 1999. A grammar of Chukchi. PhD thesis, Australian National University.
Duranti, Alessandro & Ochs, Elinor. 1990. Genitive constructions and agency in Samoan discourse.
In: Studies in Language 14, pp. 1-23.
Durie, Mark. 1985. A grammar of Acehnese (on the basis of a dialect of North Acehnese).
Dordrecht/Cinnaminson: Foris Publications.
Ebert, Karen. 1997. A grammar of Athpare. München/Newcastle: Lincom Europa.
Elifort, W illiam H., Kroeber, Paul D. & Peterson Karen L. (eds.). 1985. Papers from the parasession
on Causatives and Agentivity at the Twenty-first Regional Meeting of CLS. Chicago.
England, Nora. 1983. Ergativity in Mamean (M ayan) languages. International Journal of American
Linguistics 19, pp. 1-19.
Éva, Hompó. 1990. Grammatical relations in Gamo: a pilot sketch. In: Hayward, Richard J. (ed.),
Omotic’ language studies. University of London: SOAS, pp. 356-405.
Evans, Nicholas D. 1995. A grammar of Kayardild. Berlin/New York: Mouton de Gruyter.
Everett, Daniel L. 1986. Pirahã. In: Derbyshire & Pullum (eds.) 1986, pp. 200-325.
Everett, Daniel L. & Kern, Barbara. 1997. Wari. London: Routledge.
Farrell, Tim. 1995. Fading ergativity? A study of ergativity in Balochi. In: Bennett et al (eds.) 1995,
pp. 218-243.
Fedson, Vijayarani Jotimuttu. 1985. A note on periphrastic serial or compound verb causative
constructions in Tamil. In: Eilfort et al (eds.) 1985, pp. 13-20.
Foley, W illiam A. 1986. The Papuan languages of New Guinea. Cambridge University Press.
Foley, W illiam A. 1991. The Yimas language of New Guinea. Stanford University Press.
492

Foley, W illiam A. & Van Valin, Robert D. Jr. 1985. Information packaging in the clause. In: Shopen
(ed.) 1985, pp. 282-364.
Fortescue, Michael. 1984. West Greenlandic. London: Croom Helm.
Fox, Barbara & Hopper, Paul (eds.). 1994. Voice: Form and function. Amsterdam/Philadelphia: John
Benjamins.
Frajzyngier, Zygmunt & Curl, Traci S. (eds.). 2000. Reciprocals: forms and functions.
Amsterdam/Philadelphia: John Benjamins.
Frajzyngier, Zygmunt, Curl Traci S. (eds.) 2000b. Reflexives: forms and functions.
Amsterdam/Philadelphia: John Benjamins.
Frank, Paul. 1990. Ika syntax. Arlington: SIL International and University of Texas.
Gadelii, Karl Erland. 1997. Lesser Antillean French Creole and universal grammar. Gothenburg:
Kompendiet.
Gair, James W . 1990. Subjects, cases and INFL in Sinhala. In: Verma & Mohanan (eds.) 1990, pp.
13-41.
Gair, James W. 1998. Action involvement categories in colloquial Sinhala. In: Barbara C. Lust, (ed.).
Studies in South Asian linguistics. Oxford University Press, pp. 25-43.
Gair, James W. 1998b. Subjects, case and INFL in Sinhala (edited version of Gair 1990). In: Barbara
C. Lust, (ed.). Studies in South Asian linguistics. Oxford University Press, pp. 66-86.
Gair, James W . & Paolillo, John C. 1997. Sinhala. München/Newcastle: Lincom Europa.
Gerdts, Donna B. 1984. A relational analysis of Halkomelem causals. In Coo, E.D. & Gerdts, Donna
B. (eds.), The syntax of native American languages (Syntax and Semantics Vol. 16). New
York: Academic Press.
Givón, Talmy. 1979. On understanding grammar. New York: Academic Press.
Givón, Talmy. 1985. Ergative morphology and transitivity gradients in Newari. In: Plank (ed.) 1985,
pp. 89-107.
Givón, Talmy (ed.). 1994. Voice and inversion. Amsterdam/Philadelphia: John Benjamins.
Givón, Talmy. 1995. Functionalism and grammar. Amsterdam/Philadelphia: John Benjamins.
Givón, Talmy (ed.). 1997. Grammatical relations: A functionalist perspective.
Amsterdam/Philadelphia: John Benjamins.
Givón, Talmy. 2001. Syntax. Volume II. Amsterdam/Philadelphia: John Benjamins.
Glock, Naomi. 1972. Clause and sentence in Saramaccan. In: Journal of African Linguistics 11, pp.
45-61.
Grewendorf, Günter. 1988. Aspekte der deutschen Syntax. Eine Rektion-Bindungs-Analyse.
Tübingen: Günter Narr.
Grimes, C.E. 1991. The Buru language of eastern Indonesia. PhD thesis, Australian National
University.
Guerssel, Mohammed. 1986. On Berber verbs of change: A study of transitivity alternations. Center
for Cognitive Science, MIT, Cambridge.
Guerssel, Mohammed, Kenneth Hale, Mary Laughren, Beth Levin, and Josie W hite Eagle. 1985. A
 493

cross-linguistic study of transitivity alternations. In Eilfort et al (ed.) 1985, pp. 48-63.

Guirardello, Raquel. 1999. Trumai. In: Aikhenvald Alexandra Y. & Dixon, Robert Malcolm W ard,
Other small families and isolates (pp.341-383 of Dixon & Aikhenvald (eds) 1999), pp. 351-
353.
Haig, Helen A. 1982. Passivization in M odern W estern Armenian. In: Hopper and Thompson (eds.)
1982, pp. 161-176.
Hale, Kenneth. 1968. Preliminary remarks on W albiri grammar: II. Mimeo, MIT.
Hale, Kenneth. 1973. Person marking in W albiri. In: Stephen R. Anderson & Paul Kiparsky (eds.),
A festschrift for Morris Halle. New York: Holt, Reinehart & W inston, pp. 308-344.
Hale, Kenneth 1973b. A note on subject-object inversion on Navajo. In: Kachru, Braj B. et al, Issues
in Linguistics, Papers in honor of Henry and Renée Kahane. Urbana: University of Illinois
Press, pp. 300-309.
Halliday, M ichael. 1967. Notes on transitivity and theme in English. In: Journal of Linguistics 3, pp.
37-81.
Halliday, Michael. 1968. Notes on transitivity and theme in English 3. In: Journal of Linguistics 4,
pp. 179-215.
Harris, Alice C. 1985. The Kartvelian case. New York: Academic Press.
Harris, Alice C. 1991. Mingrelian. In: Harris, Alice C. (ed.), The indigenous languages of the
Caucasus. Volume 1: The Kartvelian languages. Delmar: Caravan Books, pp. 315-394.
Harrison, S. 1976. Mokilese reference grammar. Honolulu: University Press of Hawaii.
Hasan, Ruqaiya, Cloran, Carmel & Butt, David G (eds.). 1996. Functional descriptions: Theory in
practice. Amsterdam/Philadelphia: John Benjamins.
Haspelmath, Martin. 1987. Transitivity alternations of the anticausative type. Cologne: Institut für
Sprachwissenschaft der Universität zu Köln [Arbeitspapiere].
Haspelmath, Martin. 1990. The grammaticalization of passive morphology. In: Studies in Language
14, pp. 25-72.
Haspelmath, Martin. 1993. A grammar of Lezgian. Berlin: Mouton de Gruyter.
Haspelmath, Martin 1993b. More on the typology of inchoative/causative verb alternations. In:
Comrie & Polinsky (eds.) 1993, pp. 87-120.
Haspelmath, Martin. 1999. External possession in a European areal perspective. In: Payne & Barshi
(eds.) 1999, pp. 109-135.
Haviland, John. 1979. Guugu Yimidhirr. In: Dixon & Blake (eds.) 1979, pp. 26-180.
Heath, Jeffrey. 1977. Choctaw cases. In: BLS 3, pp. 204-213.
Heide, Ute. 1989. Zur Markierung der zentralen Partizipanten im Hausa. Köln: Arbeiten des Kölner
Universalien-Projekts. Nr. 78.
Heine, Bernd & Nurse, Derek (eds.). 2000. African languages. An introduction. Cambridge
University Press.
Helbig, Gerhard & Buscha, Joachim. 1993. Deutsche Grammatik. Ein Handbuch für den
Ausländerunterricht. Leipzig/Berlin/München: Langenscheidt Verlag Enzyklopädie.
494

Herok, Thomas. 1985. Über Sinn und Bedeutung von Prädikaten. In Plank (ed.) 1985, pp. 131-157.
Hercus, Luise A. 1994. A grammar of the Arabana-Wangkangurru Language. Lake Eyre Basin,
South Australia. Canberra: Pacific Linguistics.
Hewitt, George. 1982. ‘Anti-passive’ and ‘labile’ constructions in North Caucasian. In: General
Linguistics 22, pp. 156-171.
Hewitt, George. 1989. Abkhaz. London: Croom Helm.
Hewitt, George B. 1995. Georgian – ergative, active, or what?. In: Bennett et al (eds.) 1995, pp. 202-
217.
Hiirikoski, Juhani. 1991. Transitivity of the clause and the OVS order in Finnish: A hierarchy of
semantic roles. In: Jussi Niemi (ed.), Papers from the Eighteenth Finnish Conference of
Linguistics. Joensuu, pp. 13-29.
Hiirikoski, Juhani. 1992. The Finnish word order OVS and the English agent passive: A comparison
of textual functions. Unpublished Licentiate thesis, University of Joensuu.
Hiirikoski, Juhani. Forthcoming (2002). Transitive verbs of emotion in Finnish and English. In:
Erikoiskielet ja käännösteoria - opettamisen näkökohtia. XXII Vakki-symposium, Vaasa.
Himmelmann, Nikolaus P. & W olff, John U. 1999. Toratán (Ratahan). München/Newcastle: Lincom
Europa.
Holisky, Dee Ann & Gagua, Rusudan. 1994. Tsova-Tush (Batsbi). In: Smeets (ed.) 1994, pp. 148-
212.
Hopper, Paul J. 1985. Causes and affects. In: Eilfort et al (eds.) 1985, pp. 67-88.
Hopper, Paul J. & Thompson, Sandra A. 1980. Transitivity in grammar and discourse. Language 56,
pp. 251-299.
Hopper, Paul J. & Thompson Sandra A. (eds.). 1982. Studies in transitivity. Syntax and Semantics
Vol. 15. 1982. New York: Academic Press.
Itkonen, Esa. 1997. Maailman kielten erilaisuus ja samuus. Helsinki: Gaudeamus.
Jacobs, Peter. 1994. The inverse in Squamish. In: Givón (ed.) 1994, pp. 121-145.
Jacobsen, W esley M. 1985. Morphosyntactic transitivity and semantic markedness. In Eilfort et al
(eds.) 1985, pp. 89-104.
Job, Michael. 1985. Ergativity in Lezgian. In: Plank (ed.) 1985, pp. 159-173.
Kazenin, Konstantin. 1994. On the lexical alternation of Agent-perserving and Object-perserving
transitivity alternations. In: Nordic Journal of Linguistics 17, pp. 141-154.
Kazenin, Konstantin. 1998. On patient demotion in Lak. In: Kulikov & Vater 1998, pp. 95-115.
Keen, Sandra. 1983. Yukulta. In: Dixon & Blake (eds.) 1983, pp. 191-306.
Keenan, Edward L. 1976. Towards a universal definition of ‘Subject’. In: Li (ed.) 1976, pp. 303-333.
Keenan, Edward L. 1985. Passive in the world’s languages. In: Shopen (ed.) 1985, pp. 243-281.
Keenan, Edward L. 1987. Semantic correlates of the ergative/absolutive distinction. In Keenan,
Edward L. (ed.) Universal grammar: 15 essays. London: Croom Helm, pp. 166-194.
Keenan, Edward L. & Comrie, Bernard. 1977. Noun phrase accessibility and universal grammar. In:
Linguistic Inquiry 8, pp. 63-99.
 495

Keenan, Edward L. & Comrie, Bernard. 1979. Data on the noun phrase accessibility hierarchy. In:
Language 55, pp. 333-351.
Kegl, Judy Anne. 1985. Causative marking and the construal of agency in ASL. In: Eilfort et al (ed.)
1985, pp. 120-137.
Kemmer, Suzanne. 1993. Middle voice. Amsterdam/Philadelphia: John Benjamins.
Kibrik, Andrej A. 1993. Transitivity increase in Athabaskan languages. In: Comrie & Polinsky (eds.)
1993, pp. 47-67.
Kibrik, Andrej A. 1996. Transitivity in lexicon and grammar. In: Alexandr E. Kibrik (ed.), Godoberi.
München/Newcastle: Lincom/Europa, pp. 107-146.
Kibrik, Andrej A. 1996b. Transitivity decrease in Navajo and Athabaskan: Actor-affecting
propositional derivations. In Eloise Jelinek, Sally Midgette, Keren Rice and Leslie Saxon (eds.),
Athabaskan language studies. Essays in honor of Robert W. Young. Albuquerque: University of
New Mexico Press, pp. 259-303.
Kibrik, Alexander E. 1985. Toward a typology of ergativity. In: Nichols & W oodbury (eds.) 1985,
pp. 268-323.
Kibrik, Alexandr E. 1994. Archi. In: Smeets (ed.) 1994, pp. 297-365.
Kibrik, Alexandr E. 1997. Beyond subject and object: Toward a comprehensive relational typology.
In: Linguistic Typology 1, pp. 279-346.
Kimenyi, Alexander. 1980. A relational grammar of Kinyarwanda. Berkeley: University of California
Press.
Kittilä, Seppo. 2000. Passiivin prototyypistä. In: Anneli Pajunen (ed.), Näkökulmia kielitypologiaan.
Helsinki: Suomalaisen Kirjallisuuden Seura, pp. 286-312.
Kittilä, Seppo. 2002. Remarks on the basic transitive sentence. In: Language Sciences 24, pp. 107-
130.
Kittilä, Seppo. Submitted. A typology of involuntary agent constructions (Studies in Language).
Klaiman, Miriam H. 1987. Mechanisms of ergativity in South Asia. In Dixon (ed.) 1987, pp. 61-102.
Klaiman, Miriam H. 1988. Affectedness and control: a typology of voice systems. In: Shibatani (ed.)
1988, pp. 25-83.
Klaiman, Miriam H. 1991. Grammatical voice. Cambridge University Press.
Klamer, Marian. 1998. A grammar of Kambera. Berlin/New York: Mouton de Gruyter.
Klimov, G.A. & Alekseev, M.E. 1986. Tipologija kavkazskix jazykov. Moscow: Nauka.
Klokeid, T.J. 1976. Lardil. In Dixon (ed.) 1976, pp. 550-584.
Knight, Emily. 2001. Transitivity is disrespectful: Antipassivization as a deagentivizing/object
defocusing strategy in the avoidance register of Bunuba. Paper presented at ALT IV, Santa
Barbara, July 19, 2001.
Koptjevskaja-Tamm. Maria & Muravyova, Irina A. 1993. Alutor causatives, noun incorporation and
the Mirror Principle. In: Comrie & Polinsky (eds.) 1993, pp. 287-313.
Kozinskij, Isaac Š. 1980. Nekotorye grammati eskie universalii v podsistemax vyraženija sub’ektno-
ob’ektnyx otnošenij. Doctoral dissertation, MGU Moscow.
496

Kozinsky, Isaac Š, Nedjalkov, Vladimir P. & Polinskaja, M aria S. 1988 Antipassive in Chukchee:
oblique object, object incorporation, zero object. In: Shibatani (ed.) 1988, pp. 651-706.
Kroeber, Paul D. 1999. The Salish language family: Reconstructing syntax. Lincoln/London:
University of Nebraska Press.
Kulikov, Leonid & Vater, Heinz (eds.). 1998. Typology of verbal categories. Papers presented to
Vladimir Nedjalkov on the occasion of his 70th birthday. Tübingen: Max Niemeyer Verlag.
Lakoff, George. 1977. Linguistic gestalts. In CLS 13, pp.236-287.
Langacker, Ronald. 1990. Concept, image and symbol. The cognitive basis of grammar. Berlin/New
York: Mouton de Gruyter.
Langacker, Ronald. 1991. Foundations of cognitive grammar. Volume II: Descriptive application.
Stanford University Press.
LaPolla, Randy J. 2000. Valency-changing derivations in Dulong/Rawang. In: Dixon & Aikhenvald
(eds.) 2000a, pp. 282-311.
Launey, M. 1979. Introduction à la langue et à la littérature aztèques: 1: grammaire. Paris:
L’Harmattan.
Lazard, Gilbert. 1998. Actancy. Berlin/New York: Mouton de Gruyter.
Lazard, Gilbert. 1998b. De la transitivité restreinte à la transitivité généralisée. In: Rousseau (ed.)
1998, pp. 55-84.
Lee, Hyo Sang. 1985. Causatives in Korean and the binding hierarchy. In: Eilfort et al (eds.) 1985,
pp. 138-153.
Lehmann, Christian. 1991. Predicate classes and participation. In: Seiler & Premper (eds.) 1991, pp.
183-239.
Levin, Beth. 1993. English verb classes and alternations. A preliminary investigation.
Chicago/London: University of Chicago Press.
Levine, Robert. 1977. The Skidegate dialect of Haida. Ph.D. dissertation, Columbia University.
Li, Charles N. (ed.). 1976. Subject and topic. New York: Academic Press.
Li, Charles N. & Thompson, Sandra A. 1989 [1981]. Mandarin Chinese. A functional reference
grammar. Berkeley/Los Angeles/London: University of California Press.
Lichtenberk, Frantisek 1982. Individuation hierarchies in M anam. In Hopper and Thompson (eds.)
1982, pp. 261-276.
Lichtenberk, Frantisek. 1985. Multiple uses of reciprocal constructions. In: Australian Journal of
Linguistics 5, pp. 19-41.
Lichtenberk, Frantisek. 2000. Reciprocals without reflexives. In: Frajzyngier & Curl (eds.) 2000, pp.
31-62.
Lindvall, Ann. 1998. Transitivity in discourse. A comparision of Greek, Polish and Swedish. Lund
University Press
Loos, Eugene E. 1999. Pano. In Dixon & Aikhenvald (eds.) 1999, pp. 227-250.
Louwrens, Louis J, Ingeborg M. Kosch & Albert E. Kotzé. 1995. Northern Sotho.
München/Newcastle: Lincom Europa.
 497

Luukkainen, M atti. 1988. Von der Lexik zur Grammatik. Überlegungen zu den Begriffen Wortart,
Satzglied und Satz im Rahmen eines aktionalen Sprachmodells. Helsinki: Suomalainen
Tiedeakatemia.
Lynch, John. 1998. Pacific languages. An introduction. Honolulu: University of Hawai’i Press.
Lyons, John. 1968. Introduction to theoretical linguistics. Cambridge University Press.
Lyons, John. 1977. Semantics. Cambridge University Press.
MacDonald, Lorna. 1990. A grammar of Tauya. Berlin/New York: Mouton de Gruyter.
Magier, David. 1990. Dative/accusative subjects in M arwari. In Verma & M ohahan (eds.) 1990, pp.
213-220.
Malchukov, Andrei L. 1995. Even. München/Newcastle: Lincom Europa.
Manley, T. 1972. Outline of Sre structure. Honolulu: University of Hawai’i Press.
Manning, Christopher D. 1996. Ergativity: Argument structure and grammatical relations. Stanford:
CSLI Publications.
Maran, Raw La & Clifton, John M . 1976. The causative mechanism in Jinghpaw. In: Shibatani 1976a
(ed.), pp. 443-458.
Mardirussian, Galust. 1975. Noun incorporation in universal grammar. In: CLS 11, pp. 383-389.
Martin, Jack B. 2000. Creek voice: beyond valency. In: Dixon & Aikhenvald (eds.) 2000a, pp. 375-
403.
Martins, Silvana & Martins, Valteir. 1999. Makú. In: Dixon & Aikhenvald (eds.) 1999, pp. 251-291.
Maslova, Elena. 1993. The causative in Yukaghir. In: Comrie & Polinsky (eds.) 1993, pp. 271-285.
Maslova, Elena. 1998. A grammar of Kolyma Yukaghir. PhD thesis, University of Bielefeld.
McGregor, W illiam. 1997. Grammatical structures in noun incorporation. In: Simon-Vandenbergen
et al (eds.) 1997, pp. 141-179.
Mel’ uk, Igor. 1983. Grammatical subject and the problem of the ergative construction in Lezgian.
In: Proceedings of the second conference on the non-Slavic languages of the USSR, Folio
Slavica 5, pp. 246-293.
Mel’ uk, Igor. 1992. Toward a logical analysis of the notion ‘ergative construction’. In: Studies in
Language 16, pp. 91-138.
Menovš ikov, G.A. 1967. Grammatika jazyka aziatskix éskimosov. Moscow & Leningrad: Nauka.
Merlan, Fransesca. 1983. Ngalakan grammar, texts and vocabulary. Canberra: Pacific Linguistics.
M istry, P.J. 1976. Subject in Gujarati. An examination of verb-agreement phenomena. In Verma,
Mahindra K. (ed.), The notion of subject in South Asian languages. Madison: University of
W isconsin, pp. 240-269.
Mithun, Marianne. 1984. The evolution of noun incorporation. Language 60, pp. 847-894.
Mithun, Marianne. 1994. The implication of ergativity for a Philippine voice system. In: Fox &
Hopper (eds.) 1994, pp. 247-277.
Mithun, Marianne. 1999. The languages of the Native North America. Cambridge University Press.
Mithun, Marianne. 2000. Valency-changing derivation in Central Alaskan Yup’ik. In: Dixon &
Aikhenvald (eds.) 2000a, pp. 84-114.
498

M ohanan, K.P. & Mohanan Tara. 1990. Dative subjects in Malayalam: Semantic information in
syntax. In Verma & Mohahan (eds.) 1990, pp. 43-57.
Mohanan, Tara. 1994. Argument structure in Hindi. Stanford: CSLI Publications.
Moor, M. 1985. Studien zum Lesgischen Verb. Wiesbaden: Otto Harrassowitz.
Moreno, Juan Carlos. 1985. Anticausatives: A typological sketch. In Eilfort et el (eds.), pp. 172-181.
Morphy, Frances. 1983. Djapu. A Yolngu dialect. In: Dixon & Blake (eds.) 1983, pp. 1-190.
Mortensen, Charles Arthur. 1995. A reference grammar of Northern Embera languages. Arlington:
SIL International and University of Texas.
Mosel, Ulrike. 1991. Towards a typology of valency. In Seiler & Premper (eds.) 1991, pp. 240-251.
Mosel, Ulrike. 1991b. Abstufungen der Transitivität im Palauischen. In: Seiler & Premper (eds.)
1991, pp. 400-407.
Mosel, Ulrike & Hovdhaugen, Even. 1992. Samoan reference grammar. Oslo: Scandinavian
University Press.
Moyse-Faurie, C. 1991. Classes de verbes et variations d’actance en futunien. In: Actances 6, pp. 61-
88.
Naess, Åshild. 2000. Pileni. München/Newcastle: Lincom Europa.
N’Diaye, G. 1970. Structure de dialecte basque de Maya. The Hague/Paris: Mouton.
Nedjalkov, Igor V. & Nedjalkov, Vladimir P. 1988. Stative, resultative, passive and perfect in
Evenki. In Nedjalkov (ed.) 1988, pp. 241-257.
Nedjalkov, Vlamidir P. 1988. Typology of resultative constructions. Amsterdam/Philadelphia: John
Benjamins.
Nedjalkov, Vladimir P. & Jaxontov Sergej Je. 1988. The typology of resultative constructions. In:
Nedjalkov (ed.) 1988, pp. 3-62.
Nedjalkov, Vladimir P., Otiana, G.A. & Xolodovi . 1995. Morphological and lexical causatives in
Nivkh. In: Bennett et al (eds.) 1995, pp. 60-80.
Nichols, Johanna. 1982. Ingush transitivization and detransitivization. In: Proceedings of the Eighth
Annual Meeting of the Berkeley Linguistic Society, pp. 445-462.
Nichols, Johanna. 1986. Head-marking and dependent-marking grammar. In: Language 62, pp. 56-
119.
Nichols, Johanna. 1994. Chechen. In: Smeets (ed.) 1994, pp. 1-77.
Nichols, Johanna. 1994b. Ingush. In: Smeets (ed.) 1994, pp. 80-145.
Nichols, Johanna & W oodbury Anthony C. (eds.). 1985. Grammar inside and outside the clause.
Cambridge University Press.
Noonan, Michael. 1992. A grammar of Lango. Berlin/New York: Mouton de Gruyter.
Nowak, Elke. 1996. Transforming the images. Ergativity and transitivity in Inuktitut (Eskimo).
Berlin/New York: Mouton de Gruyter.
Olson, M. 1981. Barai clause junctures. T owards a functional theory of interclausal relations.
Unpublished PhD thesis, Australian National University.
Onishi, Masayuki. 2000. Transitivity and valency-changing derivations in M otuna. In: Dixon &
 499

Aikhenvald (eds.) 2000a, pp. 115-144.

Osumi, Midori. 1998. Tinrin grammar. Honolulu: University pf Hawai’i Press.
Palmer, Frank R. 1994. Grammatical roles and relations. Cambridge University Press.
Pandharipande, Rajeshwari. 1990. Experiencer (Dative) subjects in M arathi. In: Verma & Mohanan
(eds.) 1990, pp. 161-179.
Patz, Elizabeth. 1991. Djabugay. In: Dixon & Blake (eds.) 1991, pp. 244-347.
Payne, Doris L. 1985. Degrees of inherent transitivity in Yagua verbs. In: International Journal of
American Linguistics 51, pp. 19-37.
Payne, Doris L. & Barshi, Immanuel (eds.). 1999. External possession. Amsterdam/Philadelphia:
John Benjamins.
Payne, John R. 1985. Negation. In: Shopen (ed.) 1985, pp. 197-242.
Payne, Thomas E. 1997. Describing morphosyntax. Cambridge University Press.
Peterson, David A. 1998. Discourse-functional, historical and typological aspects of applicative
constructions. PhD thesis, University of California, Berkeley.
Peterson, David A. 1998b. The morphosyntax of transitivization in Lai (Haka Chin). In: Linguistics
of the Tibeto-Burman Area 21, pp. 87-153.
Pilot-Raichoor, C. 1991. Le badaga, langue dravidienne: description et analyse. Unpublished PhD
dissertation, University of Paris III.
Pilot-Raichoor, C. 1994. L’objet en badaga. In: Bulletin de la Société de Linguistique de Paris 89,
pp. 359-397.
Plank, Frans (ed.). 1985. Relational typology. Berlin/New York: Mouton de Gruyter.
Plungian, Vladimir A. 1993. causatives in Dogon and the overlapping of causative and passive
markers. In Dixon & Polinsky (eds.) 1993, pp. 391-396.
Poedjosoedramo, Gloria Risser. 1986. Role structure in Javanese. Linguistic Studies of Indonesian
and other languages in Indonesia. Volume 24.
Polinskaja, Maria & Nedjalkov, Vladimir P. 1987. Contrasting the absolutive in Chukchee. In: Dixon
(ed.) 1987, pp. 239-269.
Popjes, Jack & Popjes, Jo. 1986. Canela-Krahô. In: Derbyshire & Pullum (eds.) 1986, pp. 128-199.
Premper, W aldfried. 1988. Zum Problem der lexikalischen Kausation (mit Daten aus dem
Arabischen). Arbeiten des Kölner Universalienprojekts 72, pp. 28-62.
Quesada, Diego J. 2000. A grammar of Teribe. München/Newcastle: Lincom Europa.
Reed, Irene, Osahito Miyaoka, Steven Jacobson, Paschal Afcan & Michael Krauss 1997. Yup’ik
Eskimo grammar. Fairbanks: Alaska Native Language Center and Yup’ik Language
W orkshop, University of Alaska.
Reid, Nicholas. 2000. Complex verb collocations in Ngan’gityemerri: a non-derivational strategy for
encoding valency alternations. In: Dixon & Aikhenvald (eds.) 2000a, pp. 333-359.
Rice, Sally. 1987. Towards a cognitive model of transitivity. Doctoral dissertation. University of
California, San Diego.
Rodrigues, Aryon D. 1999. Tupí. In Dixon & Aikhenvald (eds.) 1999, pp. 107-124.
500

Rodrigues, Ayron D. 1999b. Macro-Jê. In Dixon & Aikhenvald 1999 (eds.), pp. 165-206.
Roldán, Mercedes. 1972. Concerning Spanish datives and possessives. In: Language Sciences 21,
pp. 27-32.
Rose, David. 1996. Pitjantjatjara Processes: An Australian experiential grammar. In: Hasan et al, pp.
287-323.
Rosen, C. & W ali K. 1988. Twin passives, inversion and multistratalism in M arathi. Cornell
University: typescript.
Rousseau, André (ed.). 1998. La transitivité. Paris: Septentrion.
Rude, Noel. 1988. Ergative, passive and antipassive in Nez Perce: A discourse perspective. In:
Shibatani (ed.) 1988, pp. 547-560.
Rude, Noel. 1997. Dative shifting and double objects in Sahaptin. In: Givón (ed.) 1997, pp. 323-349.
Rumsey, Alan. 2000. Bunuba. In Dixon & Blake (eds.) 2000, pp. 34-152.
Saksena, Anuradha. 1980. The affected agent. In: Language 56, pp. 812-826.
Samain, Didier. 1998. Hypothesè dynamique ou modèle valenciel. Quleques remarques sur
l’évolution du concept de transitivité. In: Rousseau (ed) 1998, pp. 39-52.
Sastry, G.D.P. 1984. Mishmi grammar. Mysore: Central Institute of Indian Languages.
Saulwick, Adam. 2001. Towards an intragenetic typology of benefactives in Gunwinyguan with
particular reference to Rembarnnga. A talk given at ALT IV at Santa Barbara, July 21, 2001
(handout).
Schaub, W . 1985. Babungo. London: Croom Helm..
Schebeck, Bernhard. 1976. Yuulngu. In: Dixon (ed.) 1976, pp. 352-382.
Schebeck, Bernhard. 1976b. Thangu and Atjnjamathanha. In: Dixon (ed.) 1976, pp. 516-550.
Schulze, W olfgang. 1997. Tsakhur. München/Newcastle: Lincom Europa.
Schulze-Fürhoff, W olfgang. 1994. Udi. In: Smeets (ed.) 1994, pp. 447-514
Schütz, Albert. 1985. The Fijian language. Honolulu: University of Hawai’i Press.
Seiler, Hansjakob & Premper W aldfried (eds.). 1991. Partizipation: Das sprachliche Erfassen von
Sachverhalten. Tübingen: Gunter Narr.
Seiter, W illiam J. 1980. Studies in Niuean syntax. New York/London: Garland Publishing.
Sharpe, M.C. 1976. Alawa. In: Dixon (ed.) 1976, pp. 505-515.
Shayne, Jo Anne. 1982. Some semantic aspects of Yi- and Bi- in San Carlos Apache. In Hopper and
Thompson (eds.) 1982, pp. 379-407.
Shibatani, Masayoshi (ed.). 1976a. Syntax and Semantics Vol. 6. The grammar of causative
constructions. New York: Academic Press.
Shibatani, Masayoshi. 1976b. The grammar of causative constructions: A conspectus. In: Shibatani
1976a (ed.), pp. 1-40.
Shibatani, Masayoshi. 1980. The languages of Japan. Cambridge University Press.
Shibatani, Masayoshi. 1985. Passives and related constructions. A prototype analysis. In: Language
61, pp. 821-848.
Shibatani, Masayoshi (ed.). 1988. Passive and voice. Amsterdam/Philadelphia: John Benjamins.
 501

Shopen, Timothy (ed.). 1985. Language typology and syntactic description. Volume I: Clause
structure. Cambridge University Press.
Siewierska, Anna. 1984. The passive: A comparative linguistic analysis. London: Croom Helm.
Siewierska, Anna. 1991. Functional grammar. London: Routledge.
Siewierska, Anna. 1997. The formal realization of case and agreement marking: A functional
perspective. In Simon-Vandenbergen et al (eds.) 1997, pp. 181-210.
Silverstein, Michael. 1976. Hierarchy of features and ergativity. In: Dixon 1976 (ed.), 112-171.
Simon-Vandenbergen, Anne-Marie, Kristin Davidse and Dirk Noël (eds.). 1997. Reconnecting
language. Morphology and syntax in functional perspectives. Amsterdam/Philadelphia: John
Benjamins.
Skorik, P. Ja. 1968. „ukotskij jazyk. In: Vinogradov, V.V. et al (eds.), Jazyki nadorov SSSR, Vol 5.
Leningrad: Nauka.
Slobin, Dan I. 1982. The origins of grammatical encoding of events. In Hopper and Thompson (eds.)
1982, pp. 409-422.
Smeets, Rieks (ed.). 1994. The Indigenous languages of the Caucasus. Volume 4.
Smith, Ian & Johnson, Steve 2000. Kugu Nganhcara. In: Dixon & Blake (eds.) 2000, pp. 355-489.
Song, Jae Jung. 1996. Causatives and causation. London/New York: Longman.
Song, Jae Jung. 2001. Linguistic typology. Morphology and syntax. Essex: Pearson Education
Limited.
Sugita, H. 1973. Semitransitive verbs and object incorporation in Micronesian languages. In: Oceanic
Linguistics 12, pp. 393-406.
Sumbatova, Nina R. 1993. Causative constructions in Svan: further evidence for role domination. In:
Comrie & Polinsky (eds.) 1993, pp. 253-270.
Sun, Jackson T.-S. 1998. Nominal morphology in Codeng rGyalrong.. The Bulletin of the Institute
of History and Philology, Academica Sinica 69.1, 103-149.
Svantesson, Jan-Olof. 1983. Kammu phonology and morphology. Traveaux de l’institut de
linguistique de Lund 18. Lund: University of Lund.
Talmy, Leonard. 1976. Semantic causative types. In: Shibatani 1976a (ed.), pp. 43-116.
Tamura, Suzuko. 2000. The Ainu language. Tokyo: Sanseido.
Tauberschmidt, Gerhard. 1999. A Grammar of Sinaugoro. An Austronesian language of the Central
Province of Papua New Guinea. Canberra: Pacific Linguistics.
Taylor, Charles. 1985. Nkore-Kiga. London: Croom Helm.
Tchekhoff, Claude. 1985. Aspect, transitivity, ’antipassives’ and some Australian languages. In:
Plank (ed.) 1985, pp. 359-390.
Terrill, Angela. 1998. Biri. München/Newcastle: Lincom Europa.
Testelec, Yagov G. 1998. On two parameters of transitivity. In: Kulikov & Vater (eds.) 1998, pp.
29-45.
Thomas, D.M. 1969. Chrau affixes. In: Mon-Khmer Studies 3, pp. 90-107
Thompson, Chad. 1994. Passive and inverse constructions. In: Givón (ed.) 1994, pp. 47-63.
502

Trask, R.L. 1993. A dictionary of grammatical terms in linguistics. London: Routledge.

Tsunoda, Tasaku. 1985. Remarks on transitivity. In: Journal of Linguistics 21, pp. 385-396.
Tsunoda, Tasaku. 1988. Antipassives in W arrungu and other Australian languages. In: Shibatani (ed.)
1988, pp. 595-649.
Tsunoda, Tasaku. 1994. Transitivity. Encyclopedia of Language and Linguistics, pp. 4670-4677.
Tsunoda, Tasaku. 1995 [1981]. The Djaru language of Kimberley, Western Australia. Canberra:
Pacific Linguistics.
Urban, Greg. 1985. Ergativity and accusativity in Shokleng (Gê). In: International Journal of
American Linguistics 51, pp. 164- 187.
Valenzuela, Pilar M. 1997. Basic verb types and argument structures in Shipibo-Conibo. A master
thesis presented to the department of linguistics and the Graduate School of the University of
Oregon.
Valenzuela, Pilar M . 2001. Adverbial transitivity agreement in Panoan. A talk given at ALT IV at
Santa Barbara, July 19, 2001 (handout).
Van de Kerke, Simon. 1998. Verb formation in Leko: Causatives, reflexives and reciprocals. In
Kulikov & Vater (eds.) 1998, pp. 195-203.
Van den Berg, Helma. 1995. A grammar of Hunzib. München/Newcastle: Lincom/Europa.
Van der Meer, T.H. 1985. Case marking in Suruí. In: Fortune, D.L. (ed.), Porto Velho work papers.
Brasilia: SIL, pp. 208-230.
Van Driem, George. 1993. A grammar of Dumi. Berlin/New York: Mouton de Gruyter.
Van Valin, Robert D. Jr. 1977. Ergativity and universality of subjects. In: CLS 13, pp. 689-705.
Van Valin, Robert D. Jr. 1985. Case marking and the structure of the Lakhota clause. In Nichols &
W oodbury (eds.) 1985, pp. 363-413.
Van Valin, Robert D. Jr. & LaPolla, Randy J. 1997. Syntax. Cambridge University Press.
Varamasi, Marit Kana. 1999. Grammatical relations in Bahasa Indonesia. Canberra: Pacific
Linguistics.
Vaxtin, Nikolai B. 1981. Refleksiv v éskimosskom jazyke. In: Xrakovskij (ed.), Zalogovye
konstrukcii v raznostrukturnyx jazykaz. Moscow/Leningrad: Nauka.
Verma, Mahindra K. 1990. Experiencer subjects in Bhojpuri and M agahi. In Verma & Mohanan
(eds.) 1990, pp. 85-103.
Verma, Mahindra K. & Mohanan, K.P. (eds.). 1990. Experiencer subjects in South Asian languages.
Stanford: Library of Congress Cataloging-in-Publication Data.
W atters, David. 1998. The Kham language of west-central Nepal (Takale dialect). PhD dissertation,
University of Oregon.
W atters, John R. 2000. Syntax. In: Heine & Nurse (eds.) 2000, pp. 194-230.
W ierzbicka, Anna. 1996. Semantics: Primes and universals. Oxford University Press.
W ierzbicka, Anna. 1998. Anchoring linguistic typology in universal semantic primes. In: Linguistic
Typology 2, pp. 141-194.
W ijayawardhana, G.D., Wickramasinghe, D. & Bynon, T. 1995. Passive-related constructions in
 503

colloquial Sinhala. In: Bennett et al (eds.) 1995, pp. 105-141.

W ilkins, David P. 1989. Mparntwe Arrente (Aranda): Studies in the structure and semantics of
grammar. PhD Thesis, Australian National University.
W illet, Thomas. 1981. Advancements to direct object in Tepehuan. SIL (University of North Dakota
Session) work papers 25, 59-74.
W ise, Mary Ruth. 1971. Identification of participants in discourse: a study of aspects of form and
meaning in Nomatsiguenga. Dallas: Summer Institute of Linguistics.
W oodbury, A.C. 1975. Ergativity and grammatical processes. PhD thesis, University of Chicago.
W oollams, Geoff. 1996. A Grammar of Karo Batak, Sumatra. Canberra: Pacific Linguistics.
Yallop, Colin. 1977. Alyawarra. An Aboriginal language of central Australia. Canberra: Pacific
Linguistics.
Young, Robert W . & Morgan, W illiam. 1980. The Navajo language. A grammar and colloquial
dictionary. 2nd edition. Albuquerque:University of New Mexico Press.
Zimmer, Karl E. 1976. Some constraints on Turkish causativization. In Shibatani (ed.) 1976a, pp.
399-412.